Files
windmill/backend
aa4a6ffd66 fix: track outstanding service log files on the rows themselves (#10894)
* fix: track outstanding service log files on the rows themselves

Adds `log_file.indexed_at` so the service log ingest can read outstanding rows
instead of walking a cursor over `log_ts`. A row registered after the pass had
gone by its minute was skipped for good, and no ordering fixes that — an arrival
sequence fails the same way, since a row can take a lower value and commit after
a higher one has moved the cursor past it.

The migration marks existing rows with a sentinel; the first pass returns the
ones the old cursor had not reached to the queue.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EPAP96jJNYpPQ8bpxZcU1C

* [ee] refactor: drop the claim/confirm phase from the service log ingest queue

Two states are enough: a row is outstanding or it is marked. The migration no
longer creates the index for the claim sentinel, and the sqlx cache loses the
two queries the event-time cursor used.

* [ee] fix: make re-indexing a service log file idempotent

Corrects the `init_last_log_file_sent` note: a rewritten row keeps the
`indexed_at` it had, so one the indexers already took is not offered again.

* [ee] fix: let a rebuild take the rows it covered out of the ingest queue

Adds the query that releases them; the index layout stays v4.

* [ee] fix: index the lookup a rebuild releases rows by

A rebuild takes rows out of the queue by the file it read out of the store, which
is the one lookup that arrives without a `log_ts`. The primary key is
`(hostname, log_ts)`, so nothing covered it and each batch scanned every
outstanding row — worst in exactly the state a rebuild follows. Verified at 50k
outstanding rows: sequential scan becomes an index scan.

Also records `log_file.indexed_at` in the schema reference.

* [ee] fix: treat a state handed back without its line count as behind

* [ee] fix: give the converted state a line count

* [ee] fix: keep the converted cursor from being rewound by the rebuild

* [ee] fix: inherit the legacy cursor from one source, not field by field

* [ee] fix: count a file's lines against the buffer before reading it

* [ee] fix: bound the row buffer on what it holds, not on reported counts

* [ee] fix: settle the upgrade from the store rather than from event time

* [ee] docs: describe the conversion's second half as it now works

* [ee] refactor: settle the upgrade with one rebuild instead of reconciling

The migration records existing rows as done rather than marking them with a
sentinel: the indexer puts back what the old cursor had not reached on its first
pass, which is the only place that cursor's position is known.

* [ee] fix: repair the rows the old cursor skipped instead of recording them as done

The migration marks pre-existing rows with a sentinel again, so the indexer can
tell them from rows registered since and put the window's worth back on the queue.

* [ee] fix: keep a source file whole in one partition

* [ee] revert the file-atomic partition change

* [ee] fix: dedupe the public reads, and repair an index without a cursor

* [ee] fix: repair an index whose cursor is gone, and keep what the repair found

* [ee] fix: seed a pass from both axes of what a rebuild recovered

* [ee] fix: settle the cursor on what the store holds, not on what was read

* [ee] fix: an empty rebuild must not claim ground it has not covered

* [ee] test: pin the cursor a rebuild settles on

* chore: update ee-repo-ref to bc0c7051585194474078b6c1941a3fb73893d9e5

This commit updates the EE repository reference after PR #755 was merged in windmill-ee-private.

Previous ee-repo-ref: 328f5a90afeae9c683bf3294f0d9eb293a3e1a92

New ee-repo-ref: bc0c7051585194474078b6c1941a3fb73893d9e5

Automated by sync-ee-ref workflow.

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: windmill-internal-app[bot] <windmill-internal-app[bot]@users.noreply.github.com>
2026-08-31 14:06:29 +02:00
..

Windmill Backend

This folder holds all backend components, the src/ folder only contains files used to build the "root" binary.

Components

name description
windmill-api The API server, exposing functionality to other components and the frontend
windmill-audit Contains audit functionality, allowing different components to record important actions
windmill-common Common code shared by all crates
windmill-queue Contains job & flow queuing functionality, commonly written to by the API server and read from by workers
windmill-worker The worker. Used to process and execute flows & jobs.
parsers Contains code to parse signatures in different langauges.

Compile sqlx for offline ci

cargo sqlx prepare --workspace -- --bin windmill --features enterprise