Module indexer 

SQLite-backed index for ingested log entries.

This module owns the persistent storage side of logdive: schema creation, row-level deduplication via blake3, and batched inserts of 1000 rows per transaction (per the decisions log entry dated 2026-04-19). The schema is reproduced verbatim from the project doc’s “SQLite schema” section with IF NOT EXISTS added so opening an existing database is idempotent.

Indexer is an owning handle over a rusqlite::Connection. It can be constructed against a filesystem path via Indexer::open or against an in-memory database via Indexer::open_in_memory — the latter is used by the unit tests below and will also serve ad-hoc one-shot scenarios. For read-only consumers (the HTTP API in milestone 8), [Indexer:: open_read_only] opens an existing database without the schema init or directory-creation side effects of Indexer::open, and enforces read-only semantics at the SQLite level via SQLITE_OPEN_READ_ONLY.

Timestamp NOT NULL policy

The schema declares timestamp TEXT NOT NULL, but the parser produces LogEntry::timestamp = None for lines that omit the key. Rather than fabricating a fallback (which would falsely anchor those rows to ingestion time and confuse last Nh queries), the indexer skips such rows and reports them in InsertStats::skipped_no_timestamp. This mirrors the parser’s “graceful skip” philosophy — bad data is counted and dropped, never manufactured.

Structs

Indexer

Owning handle over a SQLite connection to a logdive index.

InsertStats

Outcome of an insert batch, surfaced to the CLI for progress output (“lines ingested / lines skipped per second”, per milestone 6).

Stats

Aggregate metadata about the contents of an index.

Constants

BATCH_SIZE

Size of a single insert transaction, per the decisions log (2026-04-19: “batch insert per 1000 lines”).

Functions

db_path

Resolve the path to the index database.