Pular para conteúdo

Changelog

Every module in this repository (kiln, pgstore, mysqlstore, mssqlstore, sqlitestore, kilnotel, redisbus) is released together under the same version. The GitHub releases have the full notes.

v1.0.0-rc.1 (2026-10-09)

The release candidate for v1.0.0. The code is v0.9.0's and the API is final: unless running this candidate shows that the API has to change, v1.0.0 will be the same code, and from then on Compatibility applies.

Docs

  • A documentation site at https://kiln.rafaelaugusto.dev, in English and Brazilian Portuguese, with a guide to writing a store and a table of what each store does differently.

Upgrading

  • No schema or API changes from v0.9.0.

v0.9.0 (2026-10-09)

The last release planned to break the API before v1.0.

Breaking

  • Client.OpenBatch takes a *Batch, as StartBatch does, in place of a description and meta. It inserts the batch's jobs, continuations and nested batches and leaves the batch open, and it can open a batch nested in another through Batch.Parent. Replace OpenBatch(ctx, desc, meta) with OpenBatch(ctx, &kiln.Batch{Description: desc, Meta: meta}).

Added

  • kiln.ErrConflict, so a conflict from SetRecurring can be checked without importing driver.

Docs

  • Compatibility: what stays stable across v1, how the store interface grows, and the Go versions kiln needs.
  • The package docs name mssqlstore, and the README says that pgstore.New opens a pool of its own.

Upgrading

  • No schema changes. The only code to change is a call to OpenBatch.

v0.8.2 (2026-10-09)

Added

  • kilnotel: a kiln.jobs.failed gauge with the jobs that failed for good and wait for a requeue or a delete, the number to alert on.

Fixed

  • mysqlstore: a TxWriter insert that named a parent job and reused the unique key of that parent's parent could deadlock with Finish (error 1213, seen once in about 1,400 transactions under heavy concurrency). Inserts and Requeue now take unique keys before jobs, as Finish already did.
  • mssqlstore: under heavy concurrency a TxWriter insert could still be picked as a deadlock victim (1205) beside a Requeue. It takes the same order as MySQL now, and unique keys are claimed with a seek instead of a MERGE that scanned other transactions' keys.

Docs

  • Operating: size a kind's retries to outlast the longest outage of what it calls, and alert on kiln.jobs.failed.
  • The results of the 24 hour soak, and how to run it on a server.

Upgrading

  • No schema changes. Update every kiln module to v0.8.2 together, as usual.

v0.8.1 (2026-10-03)

Fixed

  • mysqlstore: an insert through a TxWriter that names a parent job no longer deadlocks with a Requeue of that parent. The insert locked its unique keys before its parents and the requeue locked them the other way round, so the application's transaction could fail with error 1213. Inserts now lock their parents first. mssqlstore takes the same order, which removes the plain case; under heavy concurrency SQL Server can still pick such a transaction as a deadlock victim (1205).
  • mysqlstore, mssqlstore: a child released while its parent finishes is checked against a fresh clock, so it is enqueued at once instead of waiting up to a second as scheduled.
  • dashboard: changing pages no longer fades the content out and back in.

Docs

  • Recurring jobs: the spec is evaluated in UTC unless TZ says otherwise, and occurrences overlap unless Overlap(false) is set.
  • Operating: how deploys interact with ShutdownTimeout and KillGrace, and why each service should have its own queues.

Upgrading

  • No schema changes. Update every kiln module to v0.8.1 together, as usual.

v0.8.0 (2026-10-03)

Added

  • Nested batches: Batch.AddBatch nests a batch in another, which then finishes only after every batch nested in it, and Batch.Parent nests a new batch in one that is still open. driver.NewBatch, driver.BatchQuery and driver.Batch gain Parent; driver.Batch gains Nested and NestedFinished.
  • The dashboard comes in English and Brazilian Portuguese, chosen from a picker in the header, dashboard.Options.Language or the browser's language. The catalogs live in dashboard/locales.
  • The dashboard shows a batch's parent and the batches nested in it; /api/batches?parent=ID lists them.

Fixed

  • pgstore: a limit rule written after an admission retry no longer replaces a newer rule that another insert wrote in the meantime.
  • mssqlstore: finishes running in parallel no longer leave each other's batches to the leader's sweep.

Changed

  • The rate limit docs say that the bound holds over admission times. soak reports a rate window over the bound right after a database restart or stall as a warning.

Upgrading

  • Additive schema change 006_nested_batches; v0.7 servers keep running next to v0.8 during a rolling deploy, but nest batches only once every server runs v0.8.

v0.7.0 (2026-10-03)

Added

  • Job titles: kiln.Title or a Title() method on the args names a job in the dashboard.
  • A tag filter on the dashboard's job lists, backed by driver.JobQuery.Tag.
  • Unique{Replace: true} updates a holder that hasn't started with the new args, and Unique{Debounce: d} runs the job d after the last enqueue. driver.Inserted gains Replaced.
  • Job.ParentOutputs returns the outputs of a continuation's parents that succeeded.
  • Pool.Weights shares a pool's claims between its queues by weight instead of strict order.
  • soak/, a chaos harness that kills workers and restarts the database while it checks that no job is lost.

Fixed

  • drivertest's fan-in case gives parents waiting on a busy child 15s instead of 3s, enough for SQL Server on a small CI runner.

Upgrading

  • Additive schema change 005_job_extras; v0.6 servers keep running next to v0.7 during a rolling deploy.

v0.6.0 (2026-10-02)

Added

  • Job console: Job.Logf and Job.SetProgress write log lines and a progress bar that the dashboard shows live on the job's page, grouped by attempt. Stores implement the optional driver.Console; kilntest.Result carries the lines and the progress.
  • Client.SyncRecurring keeps a group of recurring jobs in step with the code: it sets the ones it's given and removes the rest of the group. driver.Recurring gains Group.
  • A Limits page in the dashboard, and GET /api/limits: per key, the rule and how many jobs are active, throttled and holding a reserved start. Stores implement the optional driver.LimitReader.

Upgrading

  • Additive schema changes 003_console and 004_recurring_group; v0.5 servers keep running next to v0.6 during a rolling deploy.

v0.5.0 (2026-10-02)

Added

  • mssqlstore: SQL Server 2019+ and Azure SQL, on any *sql.DB from go-mssqldb, with an optional driver.Bus for wakeups. It passes the same conformance suite as the other stores; the database needs READ_COMMITTED_SNAPSHOT.

Changed

  • The stores agree on a few edge cases, each pinned by a conformance case: Series leaves out the bucket at to, a Finish outcome whose State isn't one of the five is Rejected, Output is kept only for succeeded and deleted jobs, and a limit of 0 or less counts as 1.
  • Jobs that a transactional writer's SealBatch releases are admitted by Notify, like inserted ones, instead of waiting for the sweep (pgstore, mysqlstore).
  • Close can be called more than once on every store.

Fixed

  • sqlitestore reads SQLite's error codes from any driver, so an outcome SQLite refuses is Rejected with mattn and ncruces too.
  • memstore's Sweep respects its limit, and its Tx keeps a UniqueFor window from the insert.
  • mysqlstore's Delete, Requeue and Prune count only work that committed.

v0.4.1 (2026-10-02)

Fixed

  • mysqlstore: an insert no longer waits behind a limits row that another transaction holds. Only an insert that changes that key's rule does. A held row used to stall every insert in the process, with or without a limit.
  • Delete and Finish no longer deadlock over a limits row when a job is deleted while its parent finishes (pgstore and mysqlstore).

Documentation

  • Doc comments on the store packages: memstore, drivertest, pgstore, mysqlstore and sqlitestore.
  • sqlitestore needs SQLite 3.38 or later; it was documented as 3.35.

v0.4.0 (2026-10-02)

Added

  • kilnotel module: OpenTelemetry spans from the enqueue to the handler, job metrics, and server and queue gauges.
  • driver.Bus and the redisbus module, which carries wakeups over Redis Pub/Sub so MySQL and SQLite servers in other processes start new jobs within milliseconds instead of on their next poll.
  • Job.RunAt, the time the job became due.
  • driver.TxWriter, the interface of every store's transactional writer, Notify included.
  • pgstore.Store.SQLTx for applications on database/sql with pgx's stdlib driver.
  • Doc comments on the exported API of kiln, driver, cron, dashboard, kilntest, kilnotel and redisbus.

Changed

  • Requeueing a failed, succeeded or deleted job starts it over with all of its attempts, instead of giving it one more (#5).
  • The first time a job is rescued from a dead server it goes straight back to its queue; only a job rescued twice in a row waits for its backoff (#6). driver.Orphan gains LastReason.
  • A lost claim is requeued without using up an attempt.
  • Large counts on the dashboard overview are shortened (167.4k, 1.2M) so they fit their cards.

Fixed

  • Rate-limited jobs no longer start together after admission stalls: a second GCRA state per key, admit_tat, lets Burst of them start and gives the rest new slots (#1). On PostgreSQL, admission no longer waits on limits rows, which removes the deadlocks between admit statements and Finish.
  • Limited jobs enqueued through a memstore.Tx are admitted when it commits (#2).
  • A server no longer requeues a job as lost while the claim that took it is still on its way back, which could run the job twice (#4).
  • kilntest copies RunAt, and kilnotel no longer counts a bare ErrSnoozed as a snooze.

Upgrading

  • Additive schema change 002_admit_gate: a nullable admit_tat column on the limits table. v0.3 servers keep running next to this version during a rolling deploy.

v0.3.1 (2026-09-25)

  • A server that gets no notifications from its store looks for due jobs every 100ms instead of every PollInterval, so reserved rate-limit slots, delayed jobs and retries start on time on MySQL and on PostgreSQL without a working LISTEN connection.

v0.3.0 (2026-09-24)

  • Rate limits: kiln.Limit takes Rate, Per and Burst next to Max. Start times are reserved when a job is admitted (GCRA), so a backlog is released at the configured pace and each job is written once.
  • Additive schema changes are recorded in a new schema_changes table; v0.2 servers keep running during a rolling deploy.
  • driver.InsertParams gains LimitRate, LimitPer and LimitBurst, and the conformance suite a Rate group.
  • READMEs for pgstore, mysqlstore and sqlitestore.

v0.2.0 (2026-09-24)

  • sqlitestore, for any database/sql SQLite driver.
  • mysqlstore joins the regular release.
  • Go 1.27 is the minimum version; pgx v5.11.0.
  • The compat module runs the previous release next to the new one on the same database.
  • driver.CheckInsert and driver.CheckFilter validate inputs the same way for every store.
  • Servers report the kiln version they run.

v0.1.0 (2026-09-24)

First public release: fire-and-forget, delayed and recurring jobs, continuations, flows and batches, retries, snoozes and cancellation, unique jobs and concurrency limits, transactional enqueue, the dashboard, pgstore, and the drivertest conformance suite.