NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules, #443 by repository stars
Latest stable
v0.120.6
Last release today
16 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Nearly every release is documented
notes for 58 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
4 years old
2599 releases · first in 2023
Drag across the ruler to compare, or tap a release to open it.
fix: cancelled due to concurrency should show as info log
fix: cancelled due to concurrency should show as info log
feat: pretty-print json-parseable error messages as json
fix: open side panel for standalone tasks again
fix: feature flag causing flashing
chore: lint
fix: handle error state
fix: make all cancellation logs the same color
fix: attempt to optimize concurrency queries
fix: attempt to optimize concurrency queries
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Nothing published for this version
Nothing published for this version
fix: optimize GetTenantTaskStats to try to make it less likely to seq scan v1_task
fix: optimize GetTenantTaskStats to try to make it less likely to seq scan v1_task
fix: compiler
Merge remote-tracking branch 'origin/main' into belanger/serverless-o…
Merge remote-tracking branch 'origin/main' into belanger/serverless-o…
Nothing published for this version
Nothing published for this version
Nothing published for this version
fix: more gating of matches for operator spawns
fix: more gating of matches for operator spawns
fix: rm unneeded skip
fix: oh boy....
fix: oh boy.... (#4923)
Go SDK: all required symbols are exported and deprecated constructs are no longer depended on ( #4875 ).
Hatchet v0.106.5 is a concurrency-focused release, adding support for dynamic concurrency limits, shared concurrency across workflows, and two new queue-depth concurrency strategies.
input.tier == 'premium' ? 10 : 1 (#4863, #4873).workflow_a and workflow_b to 10 concurrent runs per user total. Previously this pattern required a parent router task; shared concurrency replaces it (#4845, #4873).CANCEL_QUEUED_EXCEPT_NEWEST and CANCEL_QUEUED_EXCEPT_OLDEST: configure the queue depth for each concurrency key, keeping only the most or least recent queued item until in-progress runs complete — conceptually similar to debouncing (#4793).UpdateDurableEventLogEntriesSatisfied (#4891).UpsertDurableChildSignalCreatedEvents by @mrkaye97 in #4860go test output required by @mnafees in #4878qs in the TS SDK to fix Dependabot alerts by @mnafees in #4884UpdateDurableEventLogEntriesSatisfied by @mrkaye97 in #4891isActive using a session id instead of just a timestamp by @abelanger5 in #4916RunMany should return results in the same order of the inputs by @mnafees in #4825Full Changelog: v0.105.16...v0.106.5
Nothing published for this version
Nothing published for this version
fix: hanging from child never spawned
fix: hanging from child never spawned
refactor: rm ReExecuted field on stuct
chore: rm more dead code, rename struct
fix: greptile concern
Nothing published for this version
Nothing published for this version
Nothing published for this version
feat: allow posthog flags to target users by email / id
feat: allow posthog flags to target users by email / id (#4896)
Nothing published for this version
fix race between dag starting and terminal statuses
initial commit
nested dag
try this to fix issues
fix started at incongruity
fix race between dag starting and terminal statuses
fix for stuck dags on engine restart
ok try this
ok try this
alternative fix
conditionally add to map
fix other issues
rewrite/remove bad comments
add comment
fix not null issue
Nothing published for this version
fix(engine): prune partitions in `UpdateDurableEventLogEntriesSatisfi…
fix(engine): prune partitions in `UpdateDurableEventLogEntriesSatisfi…
Nothing published for this version
Nothing published for this version
Nothing published for this version
chore: Bump github.com/rabbitmq/amqp091-go in /examples/go/embedded (…
chore: Bump github.com/rabbitmq/amqp091-go in /examples/go/embedded (…
Nothing published for this version
Nothing published for this version
Nothing published for this version
no verbose go test output required
no verbose go test output required
remove atlas steps from CI
Nothing published for this version
feat(engine): enable durable tasks running on the dag operator to be …
feat(engine): enable durable tasks running on the dag operator to be …
Nothing published for this version
Nothing published for this version
fix for stuck dags on engine restart
fix for stuck dags on engine restart
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
Nothing published for this version
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
Nothing published for this version
Nothing published for this version
Nothing published for this version
refactor(engine): bulk operation for `UpsertDurableChildSignalCreated…
refactor(engine): bulk operation for `UpsertDurableChildSignalCreated…
Nothing published for this version
chore: add more spans to ingest
chore: add more spans to ingest
chore: decrease flush interval
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
Merge branch 'main' into repro_load_test_dag
feat: implement buffer, wire up plural queries
feat: pluralize queries
feat: implement buffer, wire up plural queries
feat: wire up env vars + config
fix: query
fix: just use env vars
fix: review comments
Hatchet v0.105.16 adds two new concurrency strategies and a batch of engine fixes around pausing, durable DAGs, and task eviction.
Hatchet v0.105.16 adds two new concurrency strategies and a batch of engine fixes around pausing, durable DAGs, and task eviction.
CANCEL_QUEUED_EXCEPT_NEWEST and CANCEL_QUEUED_EXCEPT_OLDEST: cancel queued runs in a concurrency group while keeping only the newest or oldest one (#4793).IdInsertedAt struct use int64 instead of pgtype.Timestamptz by @mrkaye97 in #4797ci param and embedded mode security check by @mnafees in #4844CANCEL_QUEUED_EXCEPT_NEWEST/CANCEL_QUEUED_EXCEPT_OLDEST concurrency strategies by @juliusgeo in #4793Full Changelog: v0.105.2...v0.105.16
Nothing published for this version
fix: Concurrency Strats migration ordering
fix: Concurrency Strats migration ordering (#4847)
Nothing published for this version
fix(engine): bind workflow event subscribe before OLAP wait
Attach the tenant stream first so chunks published during hydration are not dropped, and send a hangup instead of an empty OK when the run is already terminal.
Look up the run in v1_lookup_table and hang up from ListFinalizedWorkflowRuns so subscribe does not wait on the lagging OLAP projection.
Hangup was closing SubscribeToWorkflowEvents while out-of-order chunks were still in the reorder buffer or in flight across engine pods, which dropped stream payloads. Hold hangup until the buffer drains or a short quiet window elapses.
A per-message TTL of 0 drops fanout messages whenever the subscriber's
consumer is momentarily busy, which silently loses mid-stream chunks
during putStream bursts. Subscriber queues are exclusive + auto-delete,
so a 30s TTL bounds buildup while surviving those windows. Scheduler
partition wake-ups keep TTL 0 since their shared queue outlives its
consumer.
more traces
fix(engine): do not drop stream chunks that arrive after the terminal event
The step-run terminal event rides a separate per-msgId flush buffer, so it
can leapfrog late stream chunks. Its handler flushed buffered events and
deleted the expected-index state, so any straggler re-initialized the index
to 0, buffered against an impossible expectation, and was silently discarded
by periodicCleanup: the missing-chunk holes seen in production streams.
Keep index continuity with the -1 sentinel so post-terminal stragglers flow
through immediately, and make periodicCleanup flush leftovers instead of
discarding them.
Flushing buffered chunks past a hole when the terminal event arrived turned
late stragglers into out-of-order delivery: the straggler (no longer dropped)
was emitted after chunks with higher indexes. Instead, keep buffering on the
terminal event and let the straggler fill the hole so the buffer drains in
order. The hangup drain (holds the hangup until buffers empty) and
periodicCleanup (flushes leftovers after the grace period) still cover chunks
that never arrive.
tighten flush loop
fix(engine): do not broadcast DAG QUEUED->RUNNING as workflow-run-finished
notifyDAGsUpdated published every UpdateDAGStatuses row as
workflow-run-finished, including the QUEUED->RUNNING transition when a
DAG's first task starts. The dispatcher coerced the unknown RUNNING
status into COMPLETED, hanging up SubscribeToWorkflowEvents streams
~600ms after any DAG run started (any task with onFailure is a DAG).
Filter to terminal statuses like the task-side notifier, and drop
non-terminal statuses in the converter instead of defaulting to
COMPLETED.
Reproduce the engine hanging up stream subscribers when a DAG run starts
(onFailure makes a task a DAG) and assert late joiners still receive the
tail instead of an empty close.
fix workaround
revert other fix
tighten up some comments
generate
Co-authored-by: Julius Park juliusgeo@gmail.com
fix(engine): do not broadcast DAG QUEUED->RUNNING as workflow-run-fin…
fix(engine): do not broadcast DAG QUEUED->RUNNING as workflow-run-fin…
Nothing published for this version
fix(engine): hold buffered stream chunks on terminal event to preserv…
fix(engine): hold buffered stream chunks on terminal event to preserv…
fix(engine): do not drop stream chunks that arrive after the terminal…
fix(engine): do not drop stream chunks that arrive after the terminal…
Nothing published for this version
One column per quarter.