hyperdir 0.7.0

A container for inodes implemented as a Hyperfile
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178

//! End-to-end tests for second-level inline ("blob") files
//! (see `docs/second-level-inline.md`).
//!
//! Requires AWS credentials in the environment plus `S3_BUCKET` and
//! `S3_REGION`. Marked `#[ignore]`; run explicitly:
//!
//! ```ignore
//! source ../hyperdir-env.sh
//! cargo test --test e2e_blob -- --ignored --nocapture
//! ```
//!
//! Each run uses a unique base prefix for isolation and deletes everything
//! under it on success.

use std::time::Duration;
use aws_sdk_s3::Client;
use uuid::Uuid;
use hyperdir::{HyperDirLayout, ROOT_DIR_UUID};
use hyperdir::hyper::HyperDir;
use hyperfile::file::flags::FileFlags;
use hyperfile::file::mode::FileMode;
use hyperfile::ondisk::InodeRaw;

fn bucket() -> String {
    std::env::var("S3_BUCKET").expect("S3_BUCKET not set (source ../hyperdir-env.sh)")
}

async fn make_client() -> Client {
    let region = std::env::var("S3_REGION").expect("S3_REGION not set");
    let config = aws_config::from_env().region(aws_config::Region::new(region)).load().await;
    Client::new(&config)
}

fn dir_mode() -> FileMode { FileMode::from(0o755) }
fn file_mode() -> FileMode { FileMode::from(0o644) }

/// A regular-file inode for a blob: mode S_IFREG|0644, nlink 1, given size,
/// NOT `FLAG_INLINE` (that is the ≤80B in-inode tier).
fn file_inode(size: usize) -> InodeRaw {
    InodeRaw {
        i_mode: 0o100_644, // S_IFREG | 0644
        i_nlink: 1,
        i_size: size as u64,
        ..Default::default()
    }
}

fn io_err<E: std::fmt::Display>(e: E) -> std::io::Error {
    std::io::Error::other(e.to_string())
}

async fn object_exists(client: &Client, bucket: &str, key: &str) -> bool {
    client.head_object().bucket(bucket).key(key).send().await.is_ok()
}

async fn count_suffix(client: &Client, bucket: &str, prefix: &str, suffix: &str) -> usize {
    let mut stream = client.list_objects_v2().bucket(bucket).prefix(prefix).into_paginator().send();
    let mut n = 0;
    while let Some(page) = stream.next().await {
        let page = page.expect("list");
        for o in page.contents() {
            if o.key().map(|k| k.ends_with(suffix)).unwrap_or(false) { n += 1; }
        }
    }
    n
}

async fn purge_prefix(client: &Client, bucket: &str, prefix: &str) {
    let mut stream = client.list_objects_v2().bucket(bucket).prefix(prefix).into_paginator().send();
    let mut keys: Vec<String> = Vec::new();
    while let Some(page) = stream.next().await {
        let page = page.expect("list for cleanup");
        for o in page.contents() { if let Some(k) = o.key() { keys.push(k.to_string()); } }
    }
    for chunk in keys.chunks(1000) {
        let ids: Vec<_> = chunk.iter().map(|k| {
            aws_sdk_s3::types::ObjectIdentifier::builder().key(k).build().unwrap()
        }).collect();
        let del = aws_sdk_s3::types::Delete::builder().set_objects(Some(ids)).build().unwrap();
        let _ = client.delete_objects().bucket(bucket).delete(del).send().await;
    }
}

#[tokio::test]
#[ignore]
async fn e2e_blob_lifecycle() {
    let client = make_client().await;
    let bucket = bucket();
    let base = format!("hyperdir-e2e-blob/{}", Uuid::new_v4());
    let layout = HyperDirLayout::with_base(&base);

    let result = run(&client, &bucket, &layout).await;
    purge_prefix(&client, &bucket, &format!("{}/", base)).await;
    result.expect("e2e blob flow");
}

async fn run(client: &Client, bucket: &str, layout: &HyperDirLayout) -> std::io::Result<()> {
    let _root = HyperDir::fs_create_root(client, layout, bucket, FileFlags::rdwr(), dir_mode()).await?;

    // ---- create blob "f" (200B, above the 80B in-inode cap) ----
    let data1 = vec![0xABu8; 200];
    let (f_uuid, ulid1) = HyperDir::fs_create_blob(
        client, layout, bucket, &ROOT_DIR_UUID, "f", &file_inode(data1.len()), &data1).await?;

    let e = HyperDir::fs_resolve_entry_fast(client, layout, bucket, &ROOT_DIR_UUID, "f").await?
        .expect("f resolves");
    assert_eq!(e.uuid, f_uuid);
    assert_eq!(e.blob_ulid, Some(ulid1), "entry points at v1");
    assert_eq!(e.inline, None, "a blob is not in-inode inline");
    assert_eq!(HyperDir::fs_read_blob(client, layout, bucket, &f_uuid, ulid1).await?, data1,
               "one GET returns v1 bytes");

    // blob_ulid survives compaction (folded into the bmap value)
    let mut root = HyperDir::fs_open_root(client, layout, bucket, FileFlags::rdwr()).await?;
    let _ = root.fs_compact().await?;
    drop(root);
    let e = HyperDir::fs_resolve_entry_fast(client, layout, bucket, &ROOT_DIR_UUID, "f").await?
        .expect("f after compact");
    assert_eq!(e.blob_ulid, Some(ulid1), "blob_ulid preserved through compaction");

    // ---- write a new version (300B); latest ULID wins, old version retained ----
    let data2 = vec![0xCDu8; 300];
    let ulid2 = HyperDir::fs_write_blob(
        client, layout, bucket, &ROOT_DIR_UUID, "f", &f_uuid, &file_inode(data2.len()), &data2).await?;
    assert_ne!(ulid2, ulid1);
    let e = HyperDir::fs_resolve_entry_fast(client, layout, bucket, &ROOT_DIR_UUID, "f").await?.unwrap();
    assert_eq!(e.blob_ulid, Some(ulid2), "entry now points at v2");
    assert_eq!(HyperDir::fs_read_blob(client, layout, bucket, &f_uuid, ulid2).await?, data2);
    assert_eq!(HyperDir::fs_read_blob(client, layout, bucket, &f_uuid, ulid1).await?, data1,
               "old version still present (retention/undelete window)");

    // ---- version GC (retention 0): drop v1, keep current v2 ----
    let dropped = HyperDir::fs_gc_blob_versions(client, layout, bucket, &f_uuid, ulid2, Duration::from_millis(0)).await?;
    assert_eq!(dropped, 1, "exactly the superseded v1 reclaimed");
    assert!(HyperDir::fs_read_blob(client, layout, bucket, &f_uuid, ulid1).await.is_err(), "v1 gone");
    assert_eq!(HyperDir::fs_read_blob(client, layout, bucket, &f_uuid, ulid2).await?, data2, "v2 kept");

    // ---- spill "f" to a segmented hyperfile at the same uuid ----
    let _h = HyperDir::fs_spill_blob(
        client, layout, bucket, &ROOT_DIR_UUID, "f", &f_uuid, ulid2, FileFlags::rdwr(), file_mode()).await?;
    let e = HyperDir::fs_resolve_entry_fast(client, layout, bucket, &ROOT_DIR_UUID, "f").await?.unwrap();
    assert_eq!(e.blob_ulid, None, "f is now segmented (commit flipped blob_ulid to 0)");
    let f_prefix = format!("{}/", layout.file_key(&f_uuid));
    assert_eq!(count_suffix(client, bucket, &f_prefix, ".blob").await, 0, "blob versions cleaned after spill");
    assert!(object_exists(client, bucket, &format!("{}inode", f_prefix)).await, "segmented inode exists");

    // ---- Gap B: a stray *.blob under a referenced SEGMENTED file is reclaimed ----
    // Simulate a crashed spill that left a version behind (reuse v1's ULID string).
    let stray_key = layout.blob_key(&f_uuid, &ulid1.to_string());
    client.put_object().bucket(bucket).key(&stray_key)
        .body(aws_sdk_s3::primitives::ByteStream::from(vec![0u8; 16]))
        .send().await.map_err(io_err)?;
    assert!(object_exists(client, bucket, &stray_key).await, "stray planted");
    let _ = HyperDir::fs_gc_orphans(client, layout, bucket, Duration::from_millis(0)).await?;
    assert!(!object_exists(client, bucket, &stray_key).await, "stray blob under segmented file reclaimed (Gap B)");
    assert!(object_exists(client, bucket, &format!("{}inode", f_prefix)).await, "segmented file left intact");

    // ---- orphan reclaim of a DELETED blob (no tombstone path; orphan sweep) ----
    let data3 = vec![0x11u8; 500];
    let (g_uuid, _g_ulid) = HyperDir::fs_create_blob(
        client, layout, bucket, &ROOT_DIR_UUID, "g", &file_inode(data3.len()), &data3).await?;
    let mut root = HyperDir::fs_open_root(client, layout, bucket, FileFlags::rdwr()).await?;
    let _ = root.fs_compact().await?;
    drop(root);
    HyperDir::fs_unlink(client, layout, bucket, &ROOT_DIR_UUID, "g", &g_uuid, false, Some(Duration::from_millis(0))).await?;
    let mut root = HyperDir::fs_open_root(client, layout, bucket, FileFlags::rdwr()).await?;
    let _ = root.fs_compact().await?; // fold the PreDelete: the name is gone
    drop(root);
    assert!(HyperDir::fs_resolve_entry_fast(client, layout, bucket, &ROOT_DIR_UUID, "g").await?.is_none(),
            "g unlinked");
    let g_prefix = format!("{}/", layout.file_key(&g_uuid));
    assert!(count_suffix(client, bucket, &g_prefix, ".blob").await >= 1, "g blob still present before sweep");
    let _ = HyperDir::fs_gc_orphans(client, layout, bucket, Duration::from_millis(0)).await?;
    assert_eq!(count_suffix(client, bucket, &g_prefix, ".blob").await, 0,
               "deleted blob reclaimed by the orphan sweep");

    Ok(())
}