feldera_types/

config.rs

//! Controller configuration.
//!
//! This module defines the controller configuration structure.  The leaves of
//! this structure are individual transport-specific and data-format-specific
//! endpoint configs.  We represent these configs as opaque JSON values, so
//! that the entire configuration tree can be deserialized from a JSON file.

use crate::postprocess::PostprocessorConfig;
use crate::preprocess::PreprocessorConfig;
use crate::secret_resolver::default_secrets_directory;
use crate::transport::adhoc::AdHocInputConfig;
use crate::transport::clock::{ClockConfig, ClockTimezoneOffset};
use crate::transport::datagen::DatagenInputConfig;
use crate::transport::delta_table::{DeltaTableReaderConfig, DeltaTableWriterConfig};
use crate::transport::dynamodb::DynamoDBWriterConfig;
use crate::transport::file::{FileInputConfig, FileOutputConfig};
use crate::transport::http::{HttpInputConfig, HttpOutputConfig};
use crate::transport::iceberg::IcebergReaderConfig;
use crate::transport::kafka::{KafkaInputConfig, KafkaOutputConfig};
use crate::transport::nats::NatsInputConfig;
use crate::transport::nexmark::NexmarkInputConfig;
use crate::transport::postgres::{
    PostgresCdcReaderConfig, PostgresReaderConfig, PostgresWriterConfig,
};
use crate::transport::pubsub::PubSubInputConfig;
use crate::transport::redis::RedisOutputConfig;
use crate::transport::s3::S3InputConfig;
use crate::transport::url::UrlInputConfig;
use core::fmt;
use feldera_ir::{MirNode, MirNodeId};
use serde::de::{self, MapAccess, Visitor};
use serde::{Deserialize, Deserializer, Serialize};
use serde_json::Value as JsonValue;
use std::collections::HashMap;
use std::fmt::Display;
use std::path::Path;
use std::str::FromStr;
use std::time::Duration;
use std::{borrow::Cow, cmp::max, collections::BTreeMap};
use utoipa::ToSchema;
use utoipa::openapi::{ObjectBuilder, OneOfBuilder, Ref, RefOr, Schema, SchemaType};

pub mod dev_tweaks;
pub use dev_tweaks::DevTweaks;

const DEFAULT_MAX_PARALLEL_CONNECTOR_INIT: u64 = 10;

/// Default maximum number of updates to be kept in the output buffer.
const DEFAULT_MAX_OUTPUT_BUFFER_SIZE_RECORDS: usize = 10_000_000;

/// Default value of `ConnectorConfig::max_queued_records`.
pub const fn default_max_queued_records() -> u64 {
    1_000_000
}

pub const DEFAULT_MAX_WORKER_BATCH_SIZE: u64 = 10_000;

pub const DEFAULT_CLOCK_RESOLUTION_USECS: u64 = 1_000_000;

/// Program information included in the pipeline configuration.
#[derive(Debug, Clone, Serialize, Deserialize, ToSchema, PartialEq, Eq)]
pub struct ProgramIr {
    /// The MIR of the program.
    pub mir: HashMap<MirNodeId, MirNode>,
    /// Program schema.
    pub program_schema: serde_json::Value,
}

/// Identity of a pipeline, used to record and check ownership of an S3
/// checkpoint bucket.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct PipelineIdentity {
    /// System-generated name of the pipeline (format `pipeline-<uuid>`).
    pub name: String,
    /// Name given to the pipeline by the tenant.
    pub given_name: Option<String>,
}

impl Display for PipelineIdentity {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match &self.given_name {
            Some(given_name) => write!(f, "{given_name} (id: {})", self.name),
            None => write!(f, "{}", self.name),
        }
    }
}

/// Pipeline deployment configuration.
/// It represents configuration entries directly provided by the user
/// (e.g., runtime configuration) and entries derived from the schema
/// of the compiled program (e.g., connectors). Storage configuration,
/// if applicable, is set by the runner.
#[derive(Debug, Clone, Serialize, Deserialize, ToSchema, PartialEq)]
pub struct PipelineConfig {
    /// Global controller configuration.
    #[serde(flatten)]
    #[schema(inline)]
    pub global: RuntimeConfig,

    /// Configuration for multihost pipelines.
    ///
    /// The presence of this field indicates that the pipeline is running in
    /// multihost mode.  In the pod with ordinal 0, this triggers starting the
    /// coordinator process.  In all pods, this tells the pipeline process to
    /// await a connection from the coordinator instead of initializing the
    /// pipeline immediately.
    pub multihost: Option<MultihostConfig>,

    /// Unique system-generated name of the pipeline (format: `pipeline-<uuid>`).
    /// It is unique across all tenants and cannot be changed.
    ///
    /// The `<uuid>` is also used in the naming of various resources that back the pipeline,
    /// and as such this name is useful to find/identify corresponding resources.
    pub name: Option<String>,

    /// Name given by the tenant to the pipeline. It is only unique within the same tenant, and can
    /// be changed by the tenant when the pipeline is stopped.
    ///
    /// Given a specific tenant, it can be used to find/identify a specific pipeline of theirs.
    pub given_name: Option<String>,

    /// Configuration for persistent storage
    ///
    /// If `global.storage` is `Some(_)`, this field must be set to some
    /// [`StorageConfig`].  If `global.storage` is `None``, the pipeline ignores
    /// this field.
    #[serde(default)]
    pub storage_config: Option<StorageConfig>,

    /// Directory containing values of secrets.
    ///
    /// If this is not set, a default directory is used.
    pub secrets_dir: Option<String>,

    /// Input endpoint configuration.
    #[serde(default)]
    pub inputs: BTreeMap<Cow<'static, str>, InputEndpointConfig>,

    /// Output endpoint configuration.
    #[serde(default)]
    pub outputs: BTreeMap<Cow<'static, str>, OutputEndpointConfig>,

    /// Program information.
    #[serde(default)]
    pub program_ir: Option<ProgramIr>,
}

impl PipelineConfig {
    pub fn max_parallel_connector_init(&self) -> u64 {
        max(
            self.global
                .max_parallel_connector_init
                .unwrap_or(DEFAULT_MAX_PARALLEL_CONNECTOR_INIT),
            1,
        )
    }

    /// Extract the compiler-generated subset of the configuration (connectors
    /// and program IR) used to compute pipeline diffs.
    pub fn program_info_subset(&self) -> PipelineConfigProgramInfo {
        PipelineConfigProgramInfo {
            inputs: self.inputs.clone(),
            outputs: self.outputs.clone(),
            program_ir: self.program_ir.clone(),
        }
    }

    pub fn with_storage(self, storage: Option<(StorageConfig, StorageOptions)>) -> Self {
        let (storage_config, storage_options) = storage.unzip();
        Self {
            global: RuntimeConfig {
                storage: storage_options,
                ..self.global
            },
            storage_config,
            ..self
        }
    }

    pub fn storage(&self) -> Option<(&StorageConfig, &StorageOptions)> {
        let storage_options = self.global.storage.as_ref();
        let storage_config = self.storage_config.as_ref();
        storage_config.zip(storage_options)
    }

    /// Returns this pipeline's [`PipelineIdentity`], used to identify the
    /// pipeline that owns an S3 checkpoint bucket.
    ///
    /// Returns `None` when the pipeline has no system-generated [`name`](Self::name).
    pub fn pipeline_identity(&self) -> Option<PipelineIdentity> {
        self.name.as_ref().map(|name| PipelineIdentity {
            name: name.clone(),
            given_name: self.given_name.clone(),
        })
    }

    /// Returns `self.secrets_dir`, or the default secrets directory if it isn't
    /// set.
    pub fn secrets_dir(&self) -> &Path {
        match &self.secrets_dir {
            Some(dir) => Path::new(dir.as_str()),
            None => default_secrets_directory(),
        }
    }

    /// Abbreviated config that can be printed in the log on pipeline startup.
    pub fn display_summary(&self) -> String {
        // TODO: we may want to further abbreviate connector config.
        let summary = serde_json::json!({
            "name": self.name,
            "given_name": self.given_name,
            "global": self.global,
            "storage_config": self.storage_config,
            "secrets_dir": self.secrets_dir,
            "inputs": self.inputs,
            "outputs": self.outputs
        });

        serde_json::to_string_pretty(&summary).unwrap_or_else(|_| "{}".to_string())
    }
}

/// A subset of fields in `PipelineConfig` that are generated by the compiler.
/// These fields are shipped to the pipeline by the compilation server along with
/// the program binary.
// Note: An alternative would be to embed these fields in the program binary itself
// as static strings. This would work well for program IR, but it would require recompiling
// the program anytime a connector config changes, whereas today connector changes
// do not require recompilation.
#[derive(Default, Deserialize, Serialize, Eq, PartialEq, Debug, Clone)]
pub struct PipelineConfigProgramInfo {
    /// Input endpoint configuration.
    pub inputs: BTreeMap<Cow<'static, str>, InputEndpointConfig>,

    /// Output endpoint configuration.
    #[serde(default)]
    pub outputs: BTreeMap<Cow<'static, str>, OutputEndpointConfig>,

    /// Program information.
    #[serde(default)]
    pub program_ir: Option<ProgramIr>,
}

/// Configuration for a multihost Feldera pipeline.
///
/// This configuration is primarily for the coordinator.
#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct MultihostConfig {
    /// Number of hosts to launch.
    ///
    /// For the configuration to be truly multihost, this should be at least 2.
    /// A value of 1 still runs the multihost coordinator but it only
    /// coordinates a single host.
    pub hosts: usize,
}

impl Default for MultihostConfig {
    fn default() -> Self {
        Self { hosts: 1 }
    }
}

/// Configuration for persistent storage in a [`PipelineConfig`].
#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct StorageConfig {
    /// A directory to keep pipeline state, as a path on the filesystem of the
    /// machine or container where the pipeline will run.
    ///
    /// When storage is enabled, this directory stores the data for
    /// [StorageBackendConfig::Default].
    ///
    /// When fault tolerance is enabled, this directory stores checkpoints and
    /// the log.
    pub path: String,

    /// How to cache access to storage in this pipeline.
    #[serde(default)]
    pub cache: StorageCacheConfig,
}

impl StorageConfig {
    pub fn path(&self) -> &Path {
        Path::new(&self.path)
    }
}

/// How to cache access to storage within a Feldera pipeline.
#[derive(Copy, Clone, Default, Deserialize, Serialize, Debug, PartialEq, Eq, ToSchema)]
#[serde(rename_all = "snake_case")]
pub enum StorageCacheConfig {
    /// Use the operating system's page cache as the primary storage cache.
    ///
    /// This is the default because it currently performs better than
    /// `FelderaCache`.
    #[default]
    PageCache,

    /// Use Feldera's internal cache implementation.
    ///
    /// This is under development. It will become the default when its
    /// performance exceeds that of `PageCache`.
    FelderaCache,
}

impl StorageCacheConfig {
    #[cfg(unix)]
    pub fn to_custom_open_flags(&self) -> i32 {
        match self {
            StorageCacheConfig::PageCache => (),
            StorageCacheConfig::FelderaCache => {
                #[cfg(target_os = "linux")]
                return libc::O_DIRECT;
            }
        }
        0
    }
}

/// Storage configuration for a pipeline.
#[derive(Debug, Clone, Default, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(default)]
pub struct StorageOptions {
    /// How to connect to the underlying storage.
    pub backend: StorageBackendConfig,

    /// For a batch of data maintained as part of a persistent index during a
    /// pipeline run, the minimum estimated number of bytes to write it to
    /// storage.
    ///
    /// This is provided for debugging and fine-tuning and should ordinarily be
    /// left unset.
    ///
    /// A value of 0 will write even empty batches to storage, and nonzero
    /// values provide a threshold.  `usize::MAX` would effectively disable
    /// storage for such batches.  The default is 10,048,576 (10 MiB).
    pub min_storage_bytes: Option<usize>,

    /// For a batch of data passed through the pipeline during a single step,
    /// the minimum estimated number of bytes to write it to storage.
    ///
    /// This is provided for debugging and fine-tuning and should ordinarily be
    /// left unset.  A value of 0 will write even empty batches to storage, and
    /// nonzero values provide a threshold.  `usize::MAX`, the default,
    /// effectively disables storage for such batches.  If it is set to another
    /// value, it should ordinarily be greater than or equal to
    /// `min_storage_bytes`.
    pub min_step_storage_bytes: Option<usize>,

    /// The form of compression to use in data batches.
    ///
    /// Compression has a CPU cost but it can take better advantage of limited
    /// NVMe and network bandwidth, which means that it can increase overall
    /// performance.
    pub compression: StorageCompression,

    /// The maximum size of the in-memory storage cache, in MiB.
    ///
    /// If set, the specified cache size is spread across all the foreground and
    /// background threads. If unset, each foreground or background thread cache
    /// is limited to 256 MiB.
    pub cache_mib: Option<usize>,
}

/// Backend storage configuration.
#[derive(Debug, Clone, Default, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(tag = "name", content = "config", rename_all = "snake_case")]
pub enum StorageBackendConfig {
    /// Use the default storage configuration.
    ///
    /// This currently uses the local file system.
    #[default]
    Default,

    /// Use the local file system.
    ///
    /// This uses ordinary system file operations.
    File(Box<FileBackendConfig>),

    /// Object storage.
    Object(ObjectStorageConfig),
}

impl Display for StorageBackendConfig {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            StorageBackendConfig::Default => write!(f, "default"),
            StorageBackendConfig::File(_) => write!(f, "file"),
            StorageBackendConfig::Object(_) => write!(f, "object"),
        }
    }
}

/// Storage compression algorithm.
#[derive(Debug, Copy, Clone, Default, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "snake_case")]
pub enum StorageCompression {
    /// Use Feldera's default compression algorithm.
    ///
    /// The default may change as Feldera's performance is tuned and new
    /// algorithms are introduced.
    #[default]
    Default,

    /// Do not compress.
    None,

    /// Use [Snappy](https://en.wikipedia.org/wiki/Snappy_(compression)) compression.
    Snappy,
}

#[derive(Debug, Clone, Eq, PartialEq)]
pub enum StartFromCheckpoint {
    Latest,
    Uuid(uuid::Uuid),
}

impl ToSchema<'_> for StartFromCheckpoint {
    fn schema() -> (
        &'static str,
        utoipa::openapi::RefOr<utoipa::openapi::schema::Schema>,
    ) {
        (
            "StartFromCheckpoint",
            utoipa::openapi::RefOr::T(Schema::OneOf(
                OneOfBuilder::new()
                    .item(
                        ObjectBuilder::new()
                            .schema_type(SchemaType::String)
                            .enum_values(Some(["latest"].into_iter()))
                            .build(),
                    )
                    .item(
                        ObjectBuilder::new()
                            .schema_type(SchemaType::String)
                            .format(Some(utoipa::openapi::SchemaFormat::KnownFormat(
                                utoipa::openapi::KnownFormat::Uuid,
                            )))
                            .build(),
                    )
                    .nullable(true)
                    .build(),
            )),
        )
    }
}

impl<'de> Deserialize<'de> for StartFromCheckpoint {
    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
    where
        D: Deserializer<'de>,
    {
        struct StartFromCheckpointVisitor;

        impl<'de> Visitor<'de> for StartFromCheckpointVisitor {
            type Value = StartFromCheckpoint;

            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
                formatter.write_str("a UUID string or the string \"latest\"")
            }

            fn visit_str<E>(self, value: &str) -> Result<Self::Value, E>
            where
                E: de::Error,
            {
                if value == "latest" {
                    Ok(StartFromCheckpoint::Latest)
                } else {
                    uuid::Uuid::parse_str(value)
                        .map(StartFromCheckpoint::Uuid)
                        .map_err(|_| E::invalid_value(serde::de::Unexpected::Str(value), &self))
                }
            }
        }

        deserializer.deserialize_str(StartFromCheckpointVisitor)
    }
}

impl Serialize for StartFromCheckpoint {
    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
    where
        S: serde::Serializer,
    {
        match self {
            StartFromCheckpoint::Latest => serializer.serialize_str("latest"),
            StartFromCheckpoint::Uuid(uuid) => serializer.serialize_str(&uuid.to_string()),
        }
    }
}

#[derive(Debug, Clone, Default, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct SyncConfig {
    /// The endpoint URL for the storage service.
    ///
    /// This is typically required for custom or local S3-compatible storage providers like MinIO.
    /// Example: `http://localhost:9000`
    ///
    /// Relevant rclone config key: [`endpoint`](https://rclone.org/s3/#s3-endpoint)
    pub endpoint: Option<String>,

    /// The name of the storage bucket.
    ///
    /// This may include a path to a folder inside the bucket (e.g., `my-bucket/data`).
    pub bucket: String,

    /// The region that this bucket is in.
    ///
    /// Leave empty for Minio or the default region (`us-east-1` for AWS).
    pub region: Option<String>,

    /// The name of the cloud storage provider (e.g., `"AWS"`, `"Minio"`).
    ///
    /// Used for provider-specific behavior in rclone.
    /// If omitted, defaults to `"Other"`.
    ///
    /// See [rclone S3 provider documentation](https://rclone.org/s3/#s3-provider)
    pub provider: Option<String>,

    /// The access key used to authenticate with the storage provider.
    ///
    /// If not provided, rclone will fall back to environment-based credentials, such as
    /// `RCLONE_S3_ACCESS_KEY_ID`. In Kubernetes environments using IRSA (IAM Roles for Service Accounts),
    /// this can be left empty to allow automatic authentication via the pod's service account.
    pub access_key: Option<String>,

    /// The secret key used together with the access key for authentication.
    ///
    /// If not provided, rclone will fall back to environment-based credentials, such as
    /// `RCLONE_S3_SECRET_ACCESS_KEY`. In Kubernetes environments using IRSA (IAM Roles for Service Accounts),
    /// this can be left empty to allow automatic authentication via the pod's service account.
    pub secret_key: Option<String>,

    /// When set, the pipeline will try fetch the specified checkpoint from the
    /// object store.
    ///
    /// If `fail_if_no_checkpoint` is `true`, the pipeline will fail to initialize.
    pub start_from_checkpoint: Option<StartFromCheckpoint>,

    /// When true, the pipeline will fail to initialize if fetching the
    /// specified checkpoint fails (missing, download error).
    /// When false, the pipeline will start from scratch instead.
    ///
    /// False by default.
    #[schema(default = std::primitive::bool::default)]
    #[serde(default)]
    pub fail_if_no_checkpoint: bool,

    /// The number of file transfers to run in parallel.
    /// Default: 20
    pub transfers: Option<u8>,

    /// The number of checkers to run in parallel.
    /// Default: 20
    pub checkers: Option<u8>,

    /// Set to skip post copy check of checksums, and only check the file sizes.
    /// This can significantly improve the throughput.
    /// Defualt: false
    pub ignore_checksum: Option<bool>,

    /// Number of streams to use for multi-thread downloads.
    /// Default: 10
    pub multi_thread_streams: Option<u8>,

    /// Use multi-thread download for files above this size.
    /// Format: `[size][Suffix]` (Example: 1G, 500M)
    /// Supported suffixes: k|M|G|T
    /// Default: 100M
    pub multi_thread_cutoff: Option<String>,

    /// When true, checkpoint downloads use the maximum resources available on
    /// the host: `transfers` and `checkers` are scaled to the number of CPUs,
    /// and the download buffer is allowed to grow up to most of the available
    /// memory. This maximizes download throughput at the cost of higher CPU and
    /// memory usage during a pull.
    ///
    /// When false, downloads use the values configured via `transfers`,
    /// `checkers`, and the rclone defaults instead.
    ///
    /// Default: true
    #[schema(default = default_optimize_download_resources)]
    #[serde(default = "default_optimize_download_resources")]
    pub optimize_download_resources: bool,

    /// The number of chunks of the same file that are uploaded for multipart uploads.
    /// Default: 10
    pub upload_concurrency: Option<u8>,

    /// **Deprecated.** Use `initial=standby` when starting the pipeline instead.
    #[deprecated(note = "Use `initial=standby` when starting the pipeline instead.")]
    #[schema(default = std::primitive::bool::default)]
    #[serde(default)]
    pub standby: bool,

    /// The interval (in seconds) between each attempt to fetch the latest
    /// checkpoint from object store while in standby mode.
    ///
    /// Applies only when `start_from_checkpoint` is set to `latest`.
    ///
    /// Default: 10 seconds
    #[schema(default = default_pull_interval)]
    #[serde(default = "default_pull_interval")]
    pub pull_interval: u64,

    /// The interval (in seconds) between each push of checkpoints to object store.
    ///
    /// Default: disabled (no periodic push).
    #[serde(default)]
    pub push_interval: Option<u64>,

    /// Extra flags to pass to `rclone`.
    ///
    /// WARNING: Supplying incorrect or conflicting flags can break `rclone`.
    /// Use with caution.
    ///
    /// Refer to the docs to see the supported flags:
    /// - [Global flags](https://rclone.org/flags/)
    /// - [S3 specific flags](https://rclone.org/s3/)
    pub flags: Option<Vec<String>>,

    /// The minimum number of checkpoints to retain in object store.
    /// No checkpoints will be deleted if the total count is below this threshold.
    ///
    /// Default: 10
    #[schema(default = default_retention_min_count)]
    #[serde(default = "default_retention_min_count")]
    pub retention_min_count: u32,

    /// The minimum age (in days) a checkpoint must reach before it becomes
    /// eligible for deletion. All younger checkpoints will be preserved.
    ///
    /// Default: 30
    #[schema(default = default_retention_min_age)]
    #[serde(default = "default_retention_min_age")]
    pub retention_min_age: u32,

    /// A read-only bucket used as a fallback checkpoint source.
    ///
    /// When the pipeline has no local checkpoint and `bucket` contains no
    /// checkpoint either, it will attempt to fetch the checkpoint from this
    /// location instead.  All connection settings (`endpoint`, `region`,
    /// `provider`, `access_key`, `secret_key`) are shared with `bucket`.
    ///
    /// The pipeline **never writes** to `read_bucket`.
    ///
    /// Must point to a different location than `bucket`.
    #[serde(default)]
    pub read_bucket: Option<String>,
}

fn default_pull_interval() -> u64 {
    10
}

fn default_retention_min_count() -> u32 {
    10
}

fn default_retention_min_age() -> u32 {
    30
}

fn default_optimize_download_resources() -> bool {
    true
}

impl SyncConfig {
    pub fn validate(&self) -> Result<(), String> {
        #[allow(deprecated)]
        if self.standby {
            return Err(
                "The `standby` config field has been deprecated. Use `initial=standby` when starting the pipeline instead.".to_owned()
            );
        }

        if let Some(ref rb) = self.read_bucket
            && rb == &self.bucket
        {
            return Err(
                "invalid sync config: `read_bucket` and `bucket` must point to different locations"
                    .to_owned(),
            );
        }

        Ok(())
    }
}

/// Configuration for supplying a custom pipeline StatefulSet template via a Kubernetes ConfigMap.
///
/// Operators can provide a custom StatefulSet YAML that the Kubernetes runner will use when
/// creating pipeline StatefulSets for a pipeline. The custom template must be stored as the
/// value of a key in a ConfigMap in the same namespace as the pipeline; set `name` to the
/// ConfigMap name and `key` to the entry that contains the template.
///
/// Recommendations and requirements:
/// - **Start from the default template and modify it as needed.** The default template is present
///   in ConfigMap named as `<release-name>-pipeline-template`, with key `pipelineTemplate` in the release
///   namespace and should be used as a reference.
/// - The template must contain a valid Kubernetes `StatefulSet` manifest in YAML form. The
///   runner substitutes variables in the template before parsing; therefore the final YAML
///   must be syntactically valid.
/// - The runner performs simple string substitution for the following placeholders. Please ensure these
///   placeholders are placed at appropriate location for their semantics:
///   - `{id}`: pipeline Kubernetes name (used for object names and labels)
///   - `{namespace}`: Kubernetes namespace where the pipeline runs
///   - `{pipeline_executor_image}`: container image used to run the pipeline executor
///   - `{binary_ref}`: program binary reference passed as an argument
///   - `{program_info_ref}`: program info reference passed as an argument
///   - `{pipeline_storage_path}`: mount path for persistent pipeline storage
///   - `{storage_class_name}`: storage class name to use for PVCs (if applicable)
///   - `{deployment_id}`: UUID identifying the deployment instance
///   - `{deployment_initial}`: initial desired runtime status (e.g., `provisioning`)
///   - `{bootstrap_policy}`: bootstrap policy value when applicable
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, ToSchema)]
pub struct PipelineTemplateConfig {
    /// Name of the ConfigMap containing the pipeline template.
    pub name: String,
    /// Key in the ConfigMap containing the pipeline template.
    ///
    /// If not set, defaults to `pipelineTemplate`.
    #[schema(default = default_pipeline_template_key)]
    #[serde(default = "default_pipeline_template_key")]
    pub key: String,
}

fn default_pipeline_template_key() -> String {
    "pipelineTemplate".to_string()
}

#[derive(Debug, Clone, Default, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct ObjectStorageConfig {
    /// URL.
    ///
    /// The following URL schemes are supported:
    ///
    /// * S3:
    ///   - `s3://<bucket>/<path>`
    ///   - `s3a://<bucket>/<path>`
    ///   - `https://s3.<region>.amazonaws.com/<bucket>`
    ///   - `https://<bucket>.s3.<region>.amazonaws.com`
    ///   - `https://ACCOUNT_ID.r2.cloudflarestorage.com/bucket`
    /// * Google Cloud Storage:
    ///   - `gs://<bucket>/<path>`
    /// * Microsoft Azure Blob Storage:
    ///   - `abfs[s]://<container>/<path>` (according to [fsspec](https://github.com/fsspec/adlfs))
    ///   - `abfs[s]://<file_system>@<account_name>.dfs.core.windows.net/<path>`
    ///   - `abfs[s]://<file_system>@<account_name>.dfs.fabric.microsoft.com/<path>`
    ///   - `az://<container>/<path>` (according to [fsspec](https://github.com/fsspec/adlfs))
    ///   - `adl://<container>/<path>` (according to [fsspec](https://github.com/fsspec/adlfs))
    ///   - `azure://<container>/<path>` (custom)
    ///   - `https://<account>.dfs.core.windows.net`
    ///   - `https://<account>.blob.core.windows.net`
    ///   - `https://<account>.blob.core.windows.net/<container>`
    ///   - `https://<account>.dfs.fabric.microsoft.com`
    ///   - `https://<account>.dfs.fabric.microsoft.com/<container>`
    ///   - `https://<account>.blob.fabric.microsoft.com`
    ///   - `https://<account>.blob.fabric.microsoft.com/<container>`
    ///
    /// Settings derived from the URL will override other settings.
    pub url: String,

    /// Additional options as key-value pairs.
    ///
    /// The following keys are supported:
    ///
    /// * S3:
    ///   - `access_key_id`: AWS Access Key.
    ///   - `secret_access_key`: AWS Secret Access Key.
    ///   - `region`: Region.
    ///   - `default_region`: Default region.
    ///   - `endpoint`: Custom endpoint for communicating with S3,
    ///     e.g. `https://localhost:4566` for testing against a localstack
    ///     instance.
    ///   - `token`: Token to use for requests (passed to underlying provider).
    ///   - [Other keys](https://docs.rs/object_store/latest/object_store/aws/enum.AmazonS3ConfigKey.html#variants).
    /// * Google Cloud Storage:
    ///   - `service_account`: Path to the service account file.
    ///   - `service_account_key`: The serialized service account key.
    ///   - `google_application_credentials`: Application credentials path.
    ///   - [Other keys](https://docs.rs/object_store/latest/object_store/gcp/enum.GoogleConfigKey.html).
    /// * Microsoft Azure Blob Storage:
    ///   - `access_key`: Azure Access Key.
    ///   - `container_name`: Azure Container Name.
    ///   - `account`: Azure Account.
    ///   - `bearer_token_authorization`: Static bearer token for authorizing requests.
    ///   - `client_id`: Client ID for use in client secret or Kubernetes federated credential flow.
    ///   - `client_secret`: Client secret for use in client secret flow.
    ///   - `tenant_id`: Tenant ID for use in client secret or Kubernetes federated credential flow.
    ///   - `endpoint`: Override the endpoint for communicating with blob storage.
    ///   - [Other keys](https://docs.rs/object_store/latest/object_store/azure/enum.AzureConfigKey.html#variants).
    ///
    /// Options set through the URL take precedence over those set with these
    /// options.
    #[serde(flatten)]
    pub other_options: BTreeMap<String, String>,
}

/// Configuration for local file system access.
#[derive(Debug, Clone, Default, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(default)]
pub struct FileBackendConfig {
    /// Whether to use background threads for file I/O.
    ///
    /// Background threads should improve performance, but they can reduce
    /// performance if too few cores are available. This is provided for
    /// debugging and fine-tuning and should ordinarily be left unset.
    pub async_threads: Option<bool>,

    /// Per-I/O operation sleep duration, in milliseconds.
    ///
    /// This is for simulating slow storage devices.  Do not use this in
    /// production.
    pub ioop_delay: Option<u64>,

    /// Configuration to synchronize checkpoints to object store.
    pub sync: Option<SyncConfig>,
}

/// Global pipeline configuration settings. This is the publicly
/// exposed type for users to configure pipelines.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(default)]
pub struct RuntimeConfig {
    /// Number of DBSP worker threads.
    ///
    /// Each DBSP "foreground" worker thread is paired with a "background"
    /// thread for LSM merging, making the total number of threads twice the
    /// specified number.
    ///
    /// The typical sweet spot for the number of workers is between 4 and 16.
    /// Each worker increases overall memory consumption for data structures
    /// used during a step.
    pub workers: u16,

    /// The maximum amount of memory, in Megabytes, that the pipeline is allowed to use
    /// on each host.
    ///
    /// Setting this property activates memory pressure monitoring and backpressure
    /// mechanisms. The pipeline will track the amount of remaining memory and
    /// report the memory pressure level via the `memory_pressure` metric.
    ///
    /// As the memory pressure increases, the system will apply increasing backpressure
    /// to push state cached in memory to storage, preventing the pipeline from running
    /// out of memory at the cost of some performance degradation.
    ///
    /// It is strongly recommended to set this property to prevent the pipeline from
    /// running out of memory. The setting should not exceed the memory limit of the pipeline
    /// instance.
    ///
    /// When `max_rss_mb` is not specified but `resources.memory_mb_max` is set, the
    /// latter is used as the effective memory cap for the pipeline.
    ///
    /// See [documentation on the pipeline's memory usage](https://docs.feldera.com/operations/memory)
    /// for more details.
    pub max_rss_mb: Option<u64>,

    /// DataFusion memory pool size, in MB, shared by the ad-hoc query
    /// engine and the Delta Lake / Iceberg connectors.
    ///
    /// Carved out of `max_rss_mb` (falling back to
    /// `resources.memory_mb_max`); the remainder goes to the DBSP circuit,
    /// so the two do not double-book RAM.
    ///
    /// Unset: defaults to 5% of the effective budget, capped at 2 GB.
    /// Pipelines that don't run heavy ad-hoc / Delta / Iceberg workloads
    /// can leave this unset.
    ///
    /// Sort/aggregate-heavy ad-hoc queries (especially at high `workers`
    /// counts) should set this explicitly. An under-sized pool surfaces as
    /// `ResourcesExhausted` on the failing query — the pipeline keeps
    /// running and only that query fails.
    ///
    /// No pool limit applied if no overall budget is configured.
    ///
    /// See [documentation on the pipeline's memory usage](https://docs.feldera.com/operations/memory)
    /// for more details.
    pub datafusion_memory_mb: Option<u64>,

    /// Number of DBSP hosts.
    ///
    /// The worker threads are evenly divided among the hosts.  For single-host
    /// deployments, this should be 1 (the default).
    ///
    /// Multihost pipelines are an enterprise-only preview feature.
    pub hosts: usize,

    /// Storage configuration.
    ///
    /// - If this is `None`, the default, the pipeline's state is kept in
    ///   in-memory data-structures.  This is useful if the pipeline's state
    ///   will fit in memory and if the pipeline is ephemeral and does not need
    ///   to be recovered after a restart. The pipeline will most likely run
    ///   faster since it does not need to access storage.
    ///
    /// - If set, the pipeline's state is kept on storage.  This allows the
    ///   pipeline to work with state that will not fit into memory. It also
    ///   allows the state to be checkpointed and recovered across restarts.
    #[serde(deserialize_with = "deserialize_storage_options")]
    pub storage: Option<StorageOptions>,

    /// Fault tolerance configuration.
    #[serde(deserialize_with = "deserialize_fault_tolerance")]
    pub fault_tolerance: FtConfig,

    /// Enable CPU profiler.
    ///
    /// The default value is `true`.
    pub cpu_profiler: bool,

    /// Enable pipeline tracing.
    pub tracing: bool,

    /// Jaeger tracing endpoint to send tracing information to.
    pub tracing_endpoint_jaeger: String,

    /// Minimal input batch size.
    ///
    /// The controller delays pushing input records to the circuit until at
    /// least `min_batch_size_records` records have been received (total
    /// across all endpoints) or `max_buffering_delay_usecs` microseconds
    /// have passed since at least one input records has been buffered.
    /// Defaults to 0.
    pub min_batch_size_records: u64,

    /// Maximal delay in microseconds to wait for `min_batch_size_records` to
    /// get buffered by the controller, defaults to 0.
    pub max_buffering_delay_usecs: u64,

    /// Resource reservations and limits. This is enforced
    /// only in Feldera Cloud.
    pub resources: ResourceConfig,

    /// Real-time clock resolution in microseconds.
    ///
    /// This parameter controls the execution of queries that use the `NOW()` function.  The output of such
    /// queries depends on the real-time clock and can change over time without any external
    /// inputs.  If the query uses `NOW()`, the pipeline will update the clock value and trigger incremental
    /// recomputation at most each `clock_resolution_usecs` microseconds.  If the query does not use
    /// `NOW()`, then clock value updates are suppressed and the pipeline ignores this setting.
    ///
    /// It is set to 1 second (1,000,000 microseconds) by default.
    pub clock_resolution_usecs: Option<u64>,

    /// Fixed timezone offset for the SQL `NOW()` clock.
    ///
    /// An ISO-8601 UTC offset, for example `"+05:30"` or `"-08:00"`, that the
    /// clock connector adds to every `NOW()` value it emits, so `NOW()`
    /// returns local time in that fixed timezone instead of UTC.
    ///
    /// The offset is baked into the pipeline's checkpointed state and cannot
    /// be changed when the pipeline resumes from a checkpoint: the value from
    /// the checkpoint stays in effect, and a differing new value is ignored
    /// with a warning in the pipeline log.
    #[schema(value_type = Option<String>, example = "+05:30")]
    pub clock_timezone_offset: Option<ClockTimezoneOffset>,

    /// Optionally, a list of CPU numbers for CPUs to which the pipeline may pin
    /// its worker threads.  Specify at least twice as many CPU numbers as
    /// workers.  CPUs are generally numbered starting from 0.  The pipeline
    /// might not be able to honor CPU pinning requests.
    ///
    /// CPU pinning can make pipelines run faster and perform more consistently,
    /// as long as different pipelines running on the same machine are pinned to
    /// different CPUs.
    pub pin_cpus: Vec<usize>,

    /// Timeout in seconds for the `Provisioning` phase of the pipeline.
    /// Setting this value will override the default of the runner.
    pub provisioning_timeout_secs: Option<u64>,

    /// The maximum number of connectors initialized in parallel during pipeline
    /// startup.
    ///
    /// At startup, the pipeline must initialize all of its input and output connectors.
    /// Depending on the number and types of connectors, this can take a long time.
    /// To accelerate the process, multiple connectors are initialized concurrently.
    /// This option controls the maximum number of connectors that can be initialized
    /// in parallel.
    ///
    /// The default is 10.
    pub max_parallel_connector_init: Option<u64>,

    /// Specification of additional (sidecar) containers.
    pub init_containers: Option<serde_json::Value>,

    /// Deprecated: setting this true or false does not have an effect anymore.
    pub checkpoint_during_suspend: bool,

    /// Sets the number of available runtime threads for the http server.
    ///
    /// In most cases, this does not need to be set explicitly and
    /// the default is sufficient. Can be increased in case the
    /// pipeline HTTP API operations are a bottleneck.
    ///
    /// If not specified, the default is set to `workers`.
    pub http_workers: Option<u64>,

    /// Sets the number of available runtime threads for async IO tasks.
    ///
    /// This affects some networking and file I/O operations
    /// especially adapters and ad-hoc queries.
    ///
    /// In most cases, this does not need to be set explicitly and
    /// the default is sufficient. Can be increased in case
    /// ingress, egress or ad-hoc queries are a bottleneck.
    ///
    /// If not specified, the default is set to `workers`.
    pub io_workers: Option<u64>,

    /// Environment variables for the pipeline process.
    ///
    /// These are key-value pairs injected into the pipeline process environment.
    /// Some variable names are reserved by the platform and cannot be overridden
    /// (for example `RUST_LOG`, and variables in the `FELDERA_`,
    /// `KUBERNETES_`, and `TOKIO_` namespaces).
    #[serde(default)]
    pub env: BTreeMap<String, String>,

    /// Optional settings for tweaking Feldera internals.
    pub dev_tweaks: DevTweaks,

    /// Log filtering directives.
    ///
    /// If set to a valid [tracing-subscriber] filter, this controls the log
    /// messages emitted by the pipeline process.  Otherwise, or if the filter
    /// has invalid syntax, messages at "info" severity and higher are written
    /// to the log and all others are discarded.
    ///
    /// [tracing-subscriber]: https://docs.rs/tracing-subscriber/latest/tracing_subscriber/filter/struct.EnvFilter.html#directives
    pub logging: Option<String>,

    /// ConfigMap containing a custom pipeline template (Enterprise only).
    ///
    /// This feature is only available in Feldera Enterprise. If set, the Kubernetes runner
    /// will read the template from the specified ConfigMap and use it instead of the default
    /// StatefulSet template for the configured pipeline.
    ///
    /// check [`PipelineTemplateConfig`] documentation for details.
    pub pipeline_template_configmap: Option<PipelineTemplateConfig>,
}

/// Accepts "true" and "false" and converts them to the new format.
fn deserialize_storage_options<'de, D>(deserializer: D) -> Result<Option<StorageOptions>, D::Error>
where
    D: Deserializer<'de>,
{
    struct BoolOrStruct;

    impl<'de> Visitor<'de> for BoolOrStruct {
        type Value = Option<StorageOptions>;

        fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
            formatter.write_str("boolean or StorageOptions")
        }

        fn visit_bool<E>(self, v: bool) -> Result<Self::Value, E>
        where
            E: de::Error,
        {
            match v {
                false => Ok(None),
                true => Ok(Some(StorageOptions::default())),
            }
        }

        fn visit_unit<E>(self) -> Result<Self::Value, E>
        where
            E: de::Error,
        {
            Ok(None)
        }

        fn visit_none<E>(self) -> Result<Self::Value, E>
        where
            E: de::Error,
        {
            Ok(None)
        }

        fn visit_map<M>(self, map: M) -> Result<Option<StorageOptions>, M::Error>
        where
            M: MapAccess<'de>,
        {
            Deserialize::deserialize(de::value::MapAccessDeserializer::new(map)).map(Some)
        }
    }

    deserializer.deserialize_any(BoolOrStruct)
}

/// Accepts very old 'initial_state' and 'latest_checkpoint' as enabling fault
/// tolerance.
///
/// Accepts `null` as disabling fault tolerance.
///
/// Otherwise, deserializes [FtConfig] in the way that one might otherwise
/// expect.
fn deserialize_fault_tolerance<'de, D>(deserializer: D) -> Result<FtConfig, D::Error>
where
    D: Deserializer<'de>,
{
    struct StringOrStruct;

    impl<'de> Visitor<'de> for StringOrStruct {
        type Value = FtConfig;

        fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
            formatter.write_str("none or FtConfig or 'initial_state' or 'latest_checkpoint'")
        }

        fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
        where
            E: de::Error,
        {
            match v {
                "initial_state" | "latest_checkpoint" => Ok(FtConfig {
                    model: Some(FtModel::default()),
                    ..FtConfig::default()
                }),
                _ => Err(de::Error::invalid_value(de::Unexpected::Str(v), &self)),
            }
        }

        fn visit_unit<E>(self) -> Result<Self::Value, E>
        where
            E: de::Error,
        {
            Ok(FtConfig::default())
        }

        fn visit_none<E>(self) -> Result<Self::Value, E>
        where
            E: de::Error,
        {
            Ok(FtConfig::default())
        }

        fn visit_map<M>(self, map: M) -> Result<FtConfig, M::Error>
        where
            M: MapAccess<'de>,
        {
            Deserialize::deserialize(de::value::MapAccessDeserializer::new(map))
        }
    }

    deserializer.deserialize_any(StringOrStruct)
}

impl Default for RuntimeConfig {
    fn default() -> Self {
        Self {
            workers: 8,
            max_rss_mb: None,
            datafusion_memory_mb: None,
            hosts: 1,
            storage: Some(StorageOptions::default()),
            fault_tolerance: FtConfig::default(),
            cpu_profiler: true,
            tracing: {
                // We discovered that the jaeger crate can use up gigabytes of RAM, so it's not harmless
                // to keep it on by default.
                false
            },
            tracing_endpoint_jaeger: "127.0.0.1:6831".to_string(),
            min_batch_size_records: 0,
            max_buffering_delay_usecs: 0,
            resources: ResourceConfig::default(),
            clock_resolution_usecs: { Some(DEFAULT_CLOCK_RESOLUTION_USECS) },
            clock_timezone_offset: None,
            pin_cpus: Vec::new(),
            provisioning_timeout_secs: None,
            max_parallel_connector_init: None,
            init_containers: None,
            checkpoint_during_suspend: true,
            io_workers: None,
            http_workers: None,
            env: BTreeMap::default(),
            dev_tweaks: DevTweaks::default(),
            logging: None,
            pipeline_template_configmap: None,
        }
    }
}

/// Upper bound on the default DataFusion pool size, in MB.
///
/// Spill-to-disk handles overflow; reserving more starves the circuit.
pub const DEFAULT_DATAFUSION_MEMORY_MB_CEILING: u64 = 2048;

/// Default DataFusion pool size as a percentage of the pipeline's
/// effective memory budget. The remainder is left for the DBSP circuit.
pub const DEFAULT_DATAFUSION_MEMORY_PERCENT: u64 = 5;

impl RuntimeConfig {
    /// Pipeline's effective memory budget in MB: `max_rss_mb`, falling back
    /// to `resources.memory_mb_max` (the k8s pod limit).
    pub fn effective_memory_mb(&self) -> Option<u64> {
        self.max_rss_mb.or(self.resources.memory_mb_max)
    }

    /// Resolved DataFusion pool size in MB: explicit `datafusion_memory_mb`
    /// if set, else 5% of the effective budget capped at
    /// `DEFAULT_DATAFUSION_MEMORY_MB_CEILING`. `None` if no budget is
    /// configured.
    pub fn resolved_datafusion_memory_mb(&self) -> Option<u64> {
        if let Some(explicit) = self.datafusion_memory_mb {
            return Some(explicit);
        }
        let effective = self.effective_memory_mb()?;
        let fraction = effective * DEFAULT_DATAFUSION_MEMORY_PERCENT / 100;
        Some(fraction.min(DEFAULT_DATAFUSION_MEMORY_MB_CEILING))
    }
}

/// Fault-tolerance configuration.
///
/// The default [FtConfig] (via [FtConfig::default]) disables fault tolerance,
/// which is the configuration that one gets if [RuntimeConfig] omits fault
/// tolerance configuration.
///
/// The default value for [FtConfig::model] enables fault tolerance, as
/// `Some(FtModel::default())`.  This is the configuration that one gets if
/// [RuntimeConfig] includes a fault tolerance configuration but does not
/// specify a particular model.
#[derive(Debug, Copy, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "snake_case")]
pub struct FtConfig {
    /// Fault tolerance model to use.
    #[serde(with = "none_as_string")]
    #[serde(default = "default_model")]
    #[schema(
        schema_with = none_as_string_schema::<FtModel>,
    )]
    pub model: Option<FtModel>,

    /// Interval between automatic checkpoints, in seconds.
    ///
    /// The default is 60 seconds.  Values less than 1 or greater than 3600 will
    /// be forced into that range.
    #[serde(default = "default_checkpoint_interval_secs")]
    pub checkpoint_interval_secs: Option<u64>,
}

fn default_model() -> Option<FtModel> {
    Some(FtModel::default())
}

pub fn default_checkpoint_interval_secs() -> Option<u64> {
    Some(60)
}

impl Default for FtConfig {
    fn default() -> Self {
        Self {
            model: None,
            checkpoint_interval_secs: default_checkpoint_interval_secs(),
        }
    }
}

#[cfg(test)]
mod test {
    use super::deserialize_fault_tolerance;
    use crate::config::{
        ConnectorConfig, DEFAULT_DATAFUSION_MEMORY_MB_CEILING, FtConfig, FtModel, PipelineConfig,
        ResourceConfig, RuntimeConfig, TransportConfig,
    };
    use serde::{Deserialize, Serialize};
    use serde_json::json;

    /// A configuration written by an older version of Feldera carries no
    /// `soft_delete` field, and a connector that does not use soft deletes
    /// serializes exactly as it did before the field existed.  This keeps
    /// configurations readable and writable across versions.
    #[test]
    fn soft_delete_is_backward_compatible() {
        let config: ConnectorConfig = serde_json::from_value(json!({
            "transport": {"name": "empty_input"}
        }))
        .unwrap();
        assert!(!config.soft_delete);

        let serialized = serde_json::to_value(&config).unwrap();
        assert_eq!(serialized.get("soft_delete"), None);

        let config: ConnectorConfig = serde_json::from_value(json!({
            "transport": {"name": "empty_input"},
            "soft_delete": true
        }))
        .unwrap();
        assert!(config.soft_delete);
        assert_eq!(
            serde_json::to_value(&config).unwrap().get("soft_delete"),
            Some(&json!(true))
        );
    }

    /// Soft deletes change how input is interpreted, so flipping the flag must
    /// count as a connector change even where flow-control settings do not.
    #[test]
    fn soft_delete_invalidates_checkpointed_connector_state() {
        let config = ConnectorConfig::new(TransportConfig::EmptyInput, None);
        let mut soft_delete = config.clone();
        soft_delete.soft_delete = true;

        assert!(config.equal_for_input_checkpoint_replay(&config));
        assert!(!config.equal_for_input_checkpoint_replay(&soft_delete));
    }

    fn config_with_name(name: Option<&str>) -> PipelineConfig {
        PipelineConfig {
            global: RuntimeConfig::default(),
            multihost: None,
            name: name.map(str::to_string),
            given_name: None,
            storage_config: None,
            secrets_dir: None,
            inputs: Default::default(),
            outputs: Default::default(),
            program_ir: None,
        }
    }

    #[test]
    fn pipeline_identity_uses_system_and_given_names() {
        let mut config = config_with_name(Some("pipeline-018f6f57-4e15-7438-91af-3fd21ff2a8d3"));
        config.given_name = Some("my-pipeline".to_string());

        let metadata = config.pipeline_identity().unwrap();
        assert_eq!(
            metadata.name,
            "pipeline-018f6f57-4e15-7438-91af-3fd21ff2a8d3"
        );
        assert_eq!(metadata.given_name.as_deref(), Some("my-pipeline"));
    }

    #[test]
    fn pipeline_identity_is_none_without_system_name() {
        assert_eq!(config_with_name(None).pipeline_identity(), None);
    }

    #[test]
    fn resolved_datafusion_memory_explicit_passes_through() {
        let config = RuntimeConfig {
            max_rss_mb: Some(8_000),
            datafusion_memory_mb: Some(1_500),
            ..Default::default()
        };
        assert_eq!(config.resolved_datafusion_memory_mb(), Some(1_500));
    }

    #[test]
    fn resolved_datafusion_memory_unconfigured_returns_none() {
        let config = RuntimeConfig::default();
        assert!(config.max_rss_mb.is_none());
        assert!(config.resources.memory_mb_max.is_none());
        assert_eq!(config.resolved_datafusion_memory_mb(), None);
    }

    #[test]
    fn resolved_datafusion_memory_small_budget_scales_down() {
        // Small pipelines must provision cleanly; the default just shrinks.
        let config = RuntimeConfig {
            max_rss_mb: Some(256),
            ..Default::default()
        };
        assert_eq!(config.resolved_datafusion_memory_mb(), Some(12));

        let config = RuntimeConfig {
            max_rss_mb: Some(512),
            ..Default::default()
        };
        assert_eq!(config.resolved_datafusion_memory_mb(), Some(25));
    }

    #[test]
    fn resolved_datafusion_memory_clamps_to_ceiling_for_large_budgets() {
        // 5% of 64 GB = 3.2 GB, above the 2 GB ceiling.
        let config = RuntimeConfig {
            max_rss_mb: Some(64_000),
            ..Default::default()
        };
        assert_eq!(
            config.resolved_datafusion_memory_mb(),
            Some(DEFAULT_DATAFUSION_MEMORY_MB_CEILING),
        );
    }

    #[test]
    fn resolved_datafusion_memory_midrange_uses_five_percent() {
        // 5% of 16 GB = 800 MB, inside the clamp range.
        let config = RuntimeConfig {
            max_rss_mb: Some(16_000),
            ..Default::default()
        };
        assert_eq!(config.resolved_datafusion_memory_mb(), Some(800));
    }

    #[test]
    fn resolved_datafusion_memory_falls_back_to_resources() {
        // No max_rss_mb, but resources.memory_mb_max is set.
        let config = RuntimeConfig {
            max_rss_mb: None,
            resources: ResourceConfig {
                memory_mb_max: Some(16_000),
                ..Default::default()
            },
            ..Default::default()
        };
        assert_eq!(config.resolved_datafusion_memory_mb(), Some(800));
    }

    #[test]
    fn ft_config() {
        #[derive(Serialize, Deserialize, Default, PartialEq, Eq, Debug)]
        #[serde(default)]
        struct Wrapper {
            #[serde(deserialize_with = "deserialize_fault_tolerance")]
            config: FtConfig,
        }

        // Omitting FtConfig, or specifying null, or specifying model "none", disables fault tolerance.
        for s in [
            "{}",
            r#"{"config": null}"#,
            r#"{"config": {"model": "none"}}"#,
        ] {
            let config: Wrapper = serde_json::from_str(s).unwrap();
            assert_eq!(
                config,
                Wrapper {
                    config: FtConfig {
                        model: None,
                        checkpoint_interval_secs: Some(60)
                    }
                }
            );
        }

        // Serializing disabled FT produces explicit "none" form.
        let s = serde_json::to_string(&Wrapper {
            config: FtConfig::default(),
        })
        .unwrap();
        assert!(s.contains("\"none\""));

        // `{}` for FtConfig, or `{...}` with `model` omitted, enables fault
        // tolerance.
        for s in [r#"{"config": {}}"#, r#"{"checkpoint_interval_secs": 60}"#] {
            assert_eq!(
                serde_json::from_str::<FtConfig>(s).unwrap(),
                FtConfig {
                    model: Some(FtModel::default()),
                    checkpoint_interval_secs: Some(60)
                }
            );
        }

        // `"checkpoint_interval_secs": null` disables periodic checkpointing.
        assert_eq!(
            serde_json::from_str::<FtConfig>(r#"{"checkpoint_interval_secs": null}"#).unwrap(),
            FtConfig {
                model: Some(FtModel::default()),
                checkpoint_interval_secs: None
            }
        );
    }
}

impl FtConfig {
    pub fn is_enabled(&self) -> bool {
        self.model.is_some()
    }

    /// Returns the checkpoint interval, if fault tolerance is enabled, and
    /// otherwise `None`.
    pub fn checkpoint_interval(&self) -> Option<Duration> {
        if self.is_enabled() {
            self.checkpoint_interval_secs
                .map(|interval| Duration::from_secs(interval.clamp(1, 3600)))
        } else {
            None
        }
    }
}

/// Serde implementation for de/serializing a string into `Option<T>` where
/// `"none"` indicates `None` and any other string indicates `Some`.
///
/// This could be extended to handle non-strings by adding more forwarding
/// `visit_*` methods to the Visitor implementation.  I don't see a way to write
/// them automatically.
mod none_as_string {
    use std::marker::PhantomData;

    use serde::de::{Deserialize, Deserializer, IntoDeserializer, Visitor};
    use serde::ser::{Serialize, Serializer};

    pub(super) fn serialize<S, T>(value: &Option<T>, serializer: S) -> Result<S::Ok, S::Error>
    where
        S: Serializer,
        T: Serialize,
    {
        match value.as_ref() {
            Some(value) => value.serialize(serializer),
            None => "none".serialize(serializer),
        }
    }

    struct NoneAsString<T>(PhantomData<fn() -> T>);

    impl<'de, T> Visitor<'de> for NoneAsString<T>
    where
        T: Deserialize<'de>,
    {
        type Value = Option<T>;

        fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
            formatter.write_str("string")
        }

        fn visit_none<E>(self) -> Result<Self::Value, E>
        where
            E: serde::de::Error,
        {
            Ok(None)
        }

        fn visit_str<E>(self, value: &str) -> Result<Option<T>, E>
        where
            E: serde::de::Error,
        {
            if &value.to_ascii_lowercase() == "none" {
                Ok(None)
            } else {
                Ok(Some(T::deserialize(value.into_deserializer())?))
            }
        }
    }

    pub(super) fn deserialize<'de, D, T>(deserializer: D) -> Result<Option<T>, D::Error>
    where
        D: Deserializer<'de>,
        T: Deserialize<'de>,
    {
        deserializer.deserialize_str(NoneAsString(PhantomData))
    }
}

/// Generates an OpenAPI schema for an `Option<T>` field serialized with `none_as_string`.
/// The schema is a `oneOf` with a reference to `T`'s schema and a `"none"` string enum.
fn none_as_string_schema<'a, T: ToSchema<'a> + Default + Serialize>() -> Schema {
    Schema::OneOf(
        OneOfBuilder::new()
            .item(RefOr::Ref(Ref::new(format!(
                "#/components/schemas/{}",
                T::schema().0
            ))))
            .item(
                ObjectBuilder::new()
                    .schema_type(SchemaType::String)
                    .enum_values(Some(vec!["none"])),
            )
            .default(Some(
                serde_json::to_value(T::default()).expect("Failed to serialize default value"),
            ))
            .build(),
    )
}

/// Fault tolerance model.
///
/// The ordering is significant: we consider [Self::ExactlyOnce] to be a "higher
/// level" of fault tolerance than [Self::AtLeastOnce].
#[derive(
    Debug, Copy, Clone, Default, Eq, PartialEq, PartialOrd, Ord, Serialize, Deserialize, ToSchema,
)]
#[serde(rename_all = "snake_case")]
pub enum FtModel {
    /// Each record is output at least once.  Crashes may duplicate output, but
    /// no input or output is dropped.
    AtLeastOnce,

    /// Each record is output exactly once.  Crashes do not drop or duplicate
    /// input or output.
    #[default]
    ExactlyOnce,
}

impl FtModel {
    pub fn option_as_str(value: Option<FtModel>) -> &'static str {
        value.map_or("no", |model| model.as_str())
    }

    pub fn as_str(&self) -> &'static str {
        match self {
            FtModel::AtLeastOnce => "at_least_once",
            FtModel::ExactlyOnce => "exactly_once",
        }
    }
}

pub struct FtModelUnknown;

impl FromStr for FtModel {
    type Err = FtModelUnknown;

    fn from_str(s: &str) -> Result<Self, Self::Err> {
        match s.to_ascii_lowercase().as_str() {
            "exactly_once" => Ok(Self::ExactlyOnce),
            "at_least_once" => Ok(Self::AtLeastOnce),
            _ => Err(FtModelUnknown),
        }
    }
}

/// Describes an input connector configuration
#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct InputEndpointConfig {
    /// The name of the input stream of the circuit that this endpoint is
    /// connected to.
    pub stream: Cow<'static, str>,

    /// Connector configuration.
    #[serde(flatten)]
    pub connector_config: ConnectorConfig,
}

impl InputEndpointConfig {
    pub fn new(stream: impl Into<Cow<'static, str>>, connector_config: ConnectorConfig) -> Self {
        Self {
            stream: stream.into(),
            connector_config,
        }
    }
}

/// Deserialize the `start_after` property of a connector configuration.
/// It requires a non-standard deserialization because we want to accept
/// either a string or an array of strings.
fn deserialize_start_after<'de, D>(deserializer: D) -> Result<Option<Vec<String>>, D::Error>
where
    D: Deserializer<'de>,
{
    let value = Option::<JsonValue>::deserialize(deserializer)?;
    match value {
        Some(JsonValue::String(s)) => Ok(Some(vec![s])),
        Some(JsonValue::Array(arr)) => {
            let vec = arr
                .into_iter()
                .map(|item| {
                    item.as_str()
                        .map(|s| s.to_string())
                        .ok_or_else(|| serde::de::Error::custom("invalid 'start_after' property: expected a string, an array of strings, or null"))
                })
                .collect::<Result<Vec<String>, _>>()?;
            Ok(Some(vec))
        }
        Some(JsonValue::Null) | None => Ok(None),
        _ => Err(serde::de::Error::custom(
            "invalid 'start_after' property: expected a string, an array of strings, or null",
        )),
    }
}

#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct ConnectorConfig {
    /// Send a full snapshot of a materialized view when the connector first
    /// starts. Valid for output connectors only.
    ///
    /// When `true`, the pipeline emits the current contents of the view as the
    /// initial batch the first time the connector runs. The view must be
    /// materialized (declared with `CREATE MATERIALIZED VIEW`).
    ///
    /// The snapshot is sent exactly once per connector lifetime: it does not
    /// fire again when the pipeline resumes from a checkpoint. Modifying the
    /// connector configuration or invoking the reset API triggers a fresh
    /// snapshot when the connector supports reset (e.g., Delta Lake in
    /// `truncate` mode and Postgres).
    #[serde(default)]
    pub send_snapshot: bool,

    /// Ingest deletions as insertions, recording the original polarity in the
    /// `is_delete` metadata attribute. Valid for input connectors only.
    ///
    /// When `true`, a delete received by the connector is pushed to the table
    /// as an insertion of the same record, and the connector attaches the
    /// `is_delete` metadata attribute set to `true` to it. Insertions carry no
    /// `is_delete` attribute, so a column declared as
    /// `DEFAULT CAST(CONNECTOR_METADATA()['is_delete'] AS BOOLEAN)` is `NULL`
    /// for them. The table then contains the entire history of the input
    /// stream instead of tracking its current contents.
    ///
    /// Only tables without a primary key support this mode, since deletions in
    /// a table with a primary key delete a key rather than a record.
    ///
    /// Versions of Feldera that predate this option ignore it and apply
    /// deletions as regular deletions.
    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
    pub soft_delete: bool,

    /// Transport endpoint configuration.
    pub transport: TransportConfig,

    /// Optional preprocessor configuration
    #[serde(skip_serializing_if = "Option::is_none")]
    pub preprocessor: Option<Vec<PreprocessorConfig>>,

    /// Parser configuration.
    pub format: Option<FormatConfig>,

    /// Optional postprocessor configuration
    #[serde(skip_serializing_if = "Option::is_none")]
    pub postprocessor: Option<Vec<PostprocessorConfig>>,

    /// Name of the index that the connector is attached to.
    ///
    /// This property is valid for output connectors only.  It is used with data
    /// transports and formats that expect output updates in the form of key/value
    /// pairs, where the key typically represents a unique id associated with the
    /// table or view.
    ///
    /// To support such output formats, an output connector can be attached to an
    /// index created using the SQL CREATE INDEX statement.  An index of a table
    /// or view contains the same updates as the table or view itself, indexed by
    /// one or more key columns.
    ///
    /// See individual connector documentation for details on how they work
    /// with indexes.
    pub index: Option<String>,

    /// Output buffer configuration.
    #[serde(flatten)]
    pub output_buffer_config: OutputBufferConfig,

    /// Maximum number of records from this connector to process in a single batch.
    ///
    /// When set, this caps how many records are taken from the connector’s input
    /// buffer and pushed through the circuit at once.
    ///
    /// This is typically configured lower than `max_queued_records` to allow the
    /// connector time to restart and refill its buffer while a batch is being
    /// processed.
    ///
    /// Not all input adapters honor this limit.
    ///
    /// If this is not set, the batch size is derived from `max_worker_batch_size`.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub max_batch_size: Option<u64>,

    /// Maximum number of records processed per batch, per worker thread.
    ///
    /// When `max_batch_size` is not set, this setting is used to cap
    /// the number of records that can be taken from the connector’s input
    /// buffer and pushed through the circuit at once.  The effective batch size is computed as:
    /// `max_worker_batch_size × workers`.
    ///
    /// This provides an alternative to `max_batch_size` that automatically adjusts batch
    /// size as the number of worker threads changes to maintain constant amount of
    /// work per worker per batch.
    ///
    /// Defaults to 10,000 records per worker.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub max_worker_batch_size: Option<u64>,

    /// Backpressure threshold, in records.
    ///
    /// Maximal number of records queued by the endpoint before the endpoint
    /// is paused by the backpressure mechanism.
    ///
    /// For input endpoints, this setting bounds the number of records that have
    /// been received from the input transport but haven't yet been consumed by
    /// the circuit, since the circuit is still busy processing previous inputs.
    ///
    /// For output endpoints, this setting bounds the number of records that have
    /// been produced by the circuit but not yet sent via the output transport endpoint
    /// nor stored in the output buffer (see `enable_output_buffer`).
    ///
    /// Note that this is not a hard bound: there can be a small delay between
    /// the backpressure mechanism is triggered and the endpoint is paused, during
    /// which more data may be queued.
    ///
    /// The default is 1 million.
    #[serde(default = "default_max_queued_records")]
    pub max_queued_records: u64,

    /// Backpressure threshold, in bytes.
    ///
    /// Maximal number of bytes queued by the endpoint before the endpoint
    /// is paused by the backpressure mechanism.
    ///
    /// For input endpoints, this setting bounds the number of bytes that have
    /// been received from the input transport but haven't yet been consumed by
    /// the circuit since the circuit, since the circuit is still busy processing
    /// previous inputs.
    ///
    /// This setting is not yet implemented for output endpoints.
    ///
    /// Note that this is not a hard bound: there can be a small delay between
    /// the backpressure mechanism is triggered and the endpoint is paused, during
    /// which more data may be queued.
    ///
    /// When this is unspecified, it defaults to `1000 * max_queued_records`.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub max_queued_bytes: Option<u64>,

    /// Create connector in paused state.
    ///
    /// The default is `false`.
    #[serde(default)]
    pub paused: bool,

    /// Arbitrary user-defined text labels associated with the connector.
    ///
    /// These labels can be used in conjunction with the `start_after` property
    /// to control the start order of connectors.
    #[serde(default)]
    pub labels: Vec<String>,

    /// Start the connector after all connectors with specified labels.
    ///
    /// This property is used to control the start order of connectors.
    /// The connector will not start until all connectors with the specified
    /// labels have finished processing all inputs.
    #[serde(deserialize_with = "deserialize_start_after")]
    #[serde(default)]
    pub start_after: Option<Vec<String>>,
}

impl ConnectorConfig {
    pub fn new(transport: TransportConfig, format: Option<FormatConfig>) -> Self {
        Self {
            send_snapshot: false,
            soft_delete: false,
            transport,
            preprocessor: None,
            format,
            postprocessor: None,
            index: None,
            output_buffer_config: Default::default(),
            max_batch_size: None,
            max_worker_batch_size: None,
            max_queued_records: default_max_queued_records(),
            max_queued_bytes: None,
            paused: false,
            labels: Vec::new(),
            start_after: None,
        }
    }

    pub fn with_max_batch_size(mut self, max_batch_size: Option<u64>) -> Self {
        self.max_batch_size = max_batch_size;
        self
    }

    pub fn with_max_queued_records(mut self, max_queued_records: u64) -> Self {
        self.max_queued_records = max_queued_records;
        self
    }

    /// Compare two configs modulo the `paused` field.
    ///
    /// Used to compare checkpointed and current connector configs.
    pub fn equal_modulo_paused(&self, other: &Self) -> bool {
        let mut a = self.clone();
        let mut b = other.clone();
        a.paused = false;
        b.paused = false;
        a == b
    }

    /// Compare two input connector configs modulo fields that only affect
    /// runtime flow control and do not invalidate checkpointed connector state.
    pub fn equal_for_input_checkpoint_replay(&self, other: &Self) -> bool {
        let mut a = self.clone();
        let mut b = other.clone();
        a.normalize_for_input_checkpoint_replay();
        b.normalize_for_input_checkpoint_replay();
        a == b
    }

    fn normalize_for_input_checkpoint_replay(&mut self) {
        self.paused = false;
        self.max_batch_size = None;
        self.max_worker_batch_size = None;
        self.max_queued_records = default_max_queued_records();
        self.max_queued_bytes = None;
    }

    /// Adopt input connector settings that are safe to change while replaying
    /// checkpointed connector state.
    pub fn apply_input_checkpoint_replay_config_from(&mut self, other: &Self) {
        self.max_batch_size = other.max_batch_size;
        self.max_worker_batch_size = other.max_worker_batch_size;
        self.max_queued_records = other.max_queued_records;
        self.max_queued_bytes = other.max_queued_bytes;
    }

    /// Returns `max_queued_records` or, if it is not set, the default.
    pub fn max_queued_records(&self) -> u64 {
        self.max_queued_records
    }

    /// Returns `max_queued_bytes` or, if it is not set, the default based on
    /// `max_queued_records`.
    pub fn max_queued_bytes(&self) -> u64 {
        self.max_queued_bytes
            .unwrap_or_else(|| self.max_queued_records().saturating_mul(1000))
    }
}

#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(default)]
pub struct OutputBufferConfig {
    /// Enable output buffering.
    ///
    /// The output buffering mechanism allows decoupling the rate at which the pipeline
    /// pushes changes to the output transport from the rate of input changes.
    ///
    /// By default, output updates produced by the pipeline are pushed directly to
    /// the output transport. Some destinations may prefer to receive updates in fewer
    /// bigger batches. For instance, when writing Parquet files, producing
    /// one bigger file every few minutes is usually better than creating
    /// small files every few milliseconds.
    ///
    /// To achieve such input/output decoupling, users can enable output buffering by
    /// setting the `enable_output_buffer` flag to `true`.  When buffering is enabled, output
    /// updates produced by the pipeline are consolidated in an internal buffer and are
    /// pushed to the output transport when one of several conditions is satisfied:
    ///
    /// * data has been accumulated in the buffer for more than `max_output_buffer_time_millis`
    ///   milliseconds.
    /// * buffer size exceeds `max_output_buffer_size_records` records.
    ///
    /// This flag is `false` by default.
    // TODO: on-demand output triggered via the API.
    pub enable_output_buffer: bool,

    /// Maximum time in milliseconds data is kept in the output buffer.
    ///
    /// By default, data is kept in the buffer indefinitely until one of
    /// the other output conditions is satisfied.  When this option is
    /// set the buffer will be flushed at most every
    /// `max_output_buffer_time_millis` milliseconds.
    ///
    /// NOTE: this configuration option requires the `enable_output_buffer` flag
    /// to be set.
    pub max_output_buffer_time_millis: usize,

    /// Maximum number of updates to be kept in the output buffer.
    ///
    /// This parameter bounds the maximal size of the buffer.
    /// Note that the size of the buffer is not always equal to the
    /// total number of updates output by the pipeline. Updates to the
    /// same record can overwrite or cancel previous updates.
    ///
    /// The default is 10,000,000.
    ///
    /// NOTE: this configuration option requires the `enable_output_buffer` flag
    /// to be set.
    pub max_output_buffer_size_records: usize,
}

impl Default for OutputBufferConfig {
    fn default() -> Self {
        Self {
            enable_output_buffer: false,
            max_output_buffer_size_records: DEFAULT_MAX_OUTPUT_BUFFER_SIZE_RECORDS,
            max_output_buffer_time_millis: usize::MAX,
        }
    }
}

/// Describes an output connector configuration
#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct OutputEndpointConfig {
    /// The name of the output stream of the circuit that this endpoint is
    /// connected to.
    pub stream: Cow<'static, str>,

    /// Connector configuration.
    #[serde(flatten)]
    pub connector_config: ConnectorConfig,
}

impl OutputEndpointConfig {
    pub fn new(stream: impl Into<Cow<'static, str>>, connector_config: ConnectorConfig) -> Self {
        Self {
            stream: stream.into(),
            connector_config,
        }
    }
}

/// Transport-specific endpoint configuration passed to
/// `crate::OutputTransport::new_endpoint`
/// and `crate::InputTransport::new_endpoint`.
#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(tag = "name", content = "config", rename_all = "snake_case")]
pub enum TransportConfig {
    FileInput(FileInputConfig),
    FileOutput(FileOutputConfig),
    NatsInput(NatsInputConfig),
    KafkaInput(KafkaInputConfig),
    KafkaOutput(KafkaOutputConfig),
    PubSubInput(PubSubInputConfig),
    UrlInput(UrlInputConfig),
    S3Input(S3InputConfig),
    DeltaTableInput(DeltaTableReaderConfig),
    DeltaTableOutput(DeltaTableWriterConfig),
    // Snake case would rename "DynamoDBOutput" to `dynamo_db_output`.
    // However, DynamoDB is a single word, so override the tag to `dynamodb_output`.
    #[serde(rename = "dynamodb_output")]
    DynamoDBOutput(DynamoDBWriterConfig),
    RedisOutput(RedisOutputConfig),
    // Prevent rust from complaining about large size difference between enum variants.
    IcebergInput(Box<IcebergReaderConfig>),
    PostgresInput(PostgresReaderConfig),
    PostgresCdcInput(PostgresCdcReaderConfig),
    PostgresOutput(PostgresWriterConfig),
    Datagen(DatagenInputConfig),
    Nexmark(NexmarkInputConfig),
    /// Direct HTTP input: cannot be instantiated through API
    HttpInput(HttpInputConfig),
    /// Direct HTTP output: cannot be instantiated through API
    HttpOutput(HttpOutputConfig),
    /// Ad hoc input: cannot be instantiated through API
    AdHocInput(AdHocInputConfig),
    ClockInput(ClockConfig),
    /// Output connector that discards all data.
    NullOutput,
    /// Input connector that produces no data.
    EmptyInput,
}

impl TransportConfig {
    pub fn name(&self) -> String {
        match self {
            TransportConfig::FileInput(_) => "file_input".to_string(),
            TransportConfig::FileOutput(_) => "file_output".to_string(),
            TransportConfig::NatsInput(_) => "nats_input".to_string(),
            TransportConfig::KafkaInput(_) => "kafka_input".to_string(),
            TransportConfig::KafkaOutput(_) => "kafka_output".to_string(),
            TransportConfig::PubSubInput(_) => "pub_sub_input".to_string(),
            TransportConfig::UrlInput(_) => "url_input".to_string(),
            TransportConfig::S3Input(_) => "s3_input".to_string(),
            TransportConfig::DeltaTableInput(_) => "delta_table_input".to_string(),
            TransportConfig::DeltaTableOutput(_) => "delta_table_output".to_string(),
            TransportConfig::DynamoDBOutput(_) => "dynamodb_output".to_string(),
            TransportConfig::IcebergInput(_) => "iceberg_input".to_string(),
            TransportConfig::PostgresInput(_) => "postgres_input".to_string(),
            TransportConfig::PostgresCdcInput(_) => "postgres_cdc_input".to_string(),
            TransportConfig::PostgresOutput(_) => "postgres_output".to_string(),
            TransportConfig::Datagen(_) => "datagen".to_string(),
            TransportConfig::Nexmark(_) => "nexmark".to_string(),
            TransportConfig::HttpInput(_) => "http_input".to_string(),
            TransportConfig::HttpOutput(_) => "http_output".to_string(),
            TransportConfig::AdHocInput(_) => "adhoc_input".to_string(),
            TransportConfig::RedisOutput(_) => "redis_output".to_string(),
            TransportConfig::ClockInput(_) => "clock".to_string(),
            TransportConfig::NullOutput => "null_output".to_string(),
            TransportConfig::EmptyInput => "empty_input".to_string(),
        }
    }

    /// Returns true if the connector is transient, i.e., is created and destroyed
    /// at runtime on demand, rather than being configured as part of the pipeline.
    pub fn is_transient(&self) -> bool {
        matches!(
            self,
            TransportConfig::AdHocInput(_)
                | TransportConfig::HttpInput(_)
                | TransportConfig::HttpOutput(_)
                | TransportConfig::ClockInput(_)
        )
    }

    pub fn is_http_input(&self) -> bool {
        matches!(self, TransportConfig::HttpInput(_))
    }
}

/// Data format specification used to parse raw data received from the
/// endpoint or to encode data sent to the endpoint.
#[derive(Debug, Clone, Eq, PartialEq, Hash, Serialize, Deserialize, ToSchema)]
pub struct FormatConfig {
    /// Format name, e.g., "csv", "json", "bincode", etc.
    pub name: Cow<'static, str>,

    /// Format-specific parser or encoder configuration.
    #[serde(default)]
    #[schema(value_type = Object)]
    pub config: JsonValue,
}

#[derive(Debug, Clone, PartialEq, Default, Serialize, Deserialize, ToSchema)]
#[serde(default)]
pub struct ResourceConfig {
    /// The minimum number of CPU cores to reserve
    /// for an instance of this pipeline
    #[serde(deserialize_with = "crate::serde_via_value::deserialize")]
    pub cpu_cores_min: Option<f64>,

    /// The maximum number of CPU cores to reserve
    /// for an instance of this pipeline
    #[serde(deserialize_with = "crate::serde_via_value::deserialize")]
    pub cpu_cores_max: Option<f64>,

    /// The minimum memory in Megabytes to reserve
    /// for an instance of this pipeline
    pub memory_mb_min: Option<u64>,

    /// The maximum memory in Megabytes to reserve
    /// for an instance of this pipeline
    pub memory_mb_max: Option<u64>,

    /// The total storage in Megabytes to reserve
    /// for an instance of this pipeline
    pub storage_mb_max: Option<u64>,

    /// Storage class to use for an instance of this pipeline.
    /// The class determines storage performance such as IOPS and throughput.
    pub storage_class: Option<String>,

    /// Kubernetes service account name to use for an instance of this pipeline.
    /// The account determines permissions and access controls.
    pub service_account_name: Option<String>,

    /// Kubernetes namespace to use for an instance of this pipeline.
    /// The namespace determines the scope of names for resources created
    /// for the pipeline.
    /// If not set, the pipeline will be deployed in the same namespace
    /// as the control-plane.
    // The type of this field should not be backward incompatibly changed, and its location in the
    // runtime configuration JSON (`runtime_config.resources.namespace`) should not be changed.
    pub namespace: Option<String>,
}