diff --git a/docker-compose.published.yml b/docker-compose.published.yml index e105ba82e..b4778ccaa 100644 --- a/docker-compose.published.yml +++ b/docker-compose.published.yml @@ -16,7 +16,7 @@ name: durable-workflow-server # release whose default names another image). Set DW_SERVER_IMAGE to run # any other native image, by tag or digest; pin a digest for production # change control. -x-server-image: &server-image ${DW_SERVER_IMAGE:-durableworkflow/server:3.12.0} +x-server-image: &server-image ${DW_SERVER_IMAGE:-durableworkflow/server:3.13.0} services: server: diff --git a/docs/releases/3.13.0.md b/docs/releases/3.13.0.md new file mode 100644 index 000000000..ce32937a2 --- /dev/null +++ b/docs/releases/3.13.0.md @@ -0,0 +1,85 @@ +# Durable Workflow Server 3.13.0 (native) + +Server 3.13.0 lets fan-outs within the documented completion size complete, +gives schedule-triggered runs embedded mode's deterministic ids, expires stale +worker registrations and pages the worker list, reports storage permission +failures as non-retryable, and bounds identifier fields at their column sizes. +Its immutable images are `durableworkflow/server:3.13.0` and +`ghcr.io/durable-workflow/server:3.13.0`; the published Compose default selects +3.13.0. Rolling aliases move separately after published-image qualification. + +**Upgrade:** a drop-in replacement for 3.12.x. There is no schema change and no +`schema-upgrade` step; rolling back to 3.12.0 is an ordinary image rollback. +Worker protocol and control-plane versions are unchanged. + +## Large fan-outs within the documented limits complete + +A workflow-task completion up to the advertised +`limits.max_workflow_task_completion_bytes` (32 MiB) could still be refused with +`409 history_transaction_limit_exceeded`, blocking the run before any activity +was scheduled. Native kept PHP's 16 MiB default for the per-transaction history +budget, and each `ActivityScheduled` event repeats its arguments, so a +`1,000 × 22 KB` `all()` was refused at about 30 MB. + +When `DW_V2_LIMIT_HISTORY_TRANSACTION_BYTES` is unset, the budget now defaults to +four times the completion cap (128 MiB with the default cap, never below +16 MiB). An explicitly configured value is still honored. The `409` refusal now +names `limit_kind`, `current_value` and `configured_limit`. Python SDK 2.18.0 +fails the run once with a structural-limit failure when an explicitly lowered +budget is still exceeded (durable-workflow/.github#228, #1231). + +On MariaDB, completions over about 16 MiB also need `max_allowed_packet` of at +least 64 MiB (MySQL 8 and PostgreSQL defaults are sufficient); see #1234. + +## Deterministic run ids for schedule occurrences + +A scheduled, backfilled or drained buffered occurrence now starts its run under +the same id embedded Workflow uses: + +```text +schedule.ns:{first 32 hex of sha256(namespace)}:{schedule id byte length}:{schedule id}:backfill:{occurrence Unix seconds} +``` + +When the id would exceed 191 characters, or the schedule id contains characters +outside `[A-Za-z0-9._:-]`, the schedule-id segment becomes `h:{sha256(schedule id)}`. +A repeated fire of an occurrence whose run already exists records +`already_fired_occurrence` and starts nothing. Manual triggers keep generated +ids, and existing runs keep their ids (#599, #1230). + +## Worker registration retention and a paged worker list + +A registration with no heartbeat for `DW_WORKER_REGISTRATION_RETENTION_SECONDS` +(default 86400, never shorter than the stale window) is removed, unless it still +holds an unexpired task lease. Active and draining workers heartbeat and are +never removed. Removal retires the registration's incarnation, as deregistration +does; runs, pinned build ids and pending tasks are untouched, and a build whose +last worker expired still reports `no_compatible_worker`. + +`GET /api/workers` returns at most 100 registrations per page, ordered by worker +id, with `page_size` (1–100) and `next_page_token`. The response adds +`registration_retention_seconds`. Clients that ignore `next_page_token` see the +first 100 workers of a namespace (#553, #1232). + +## Storage permission failures are access denied + +A local external-payload storage permission refusal (EACCES or EPERM) now returns +`external_payload_storage_access_denied` (HTTP 424, `retryable: false`, a fixed +message and no storage path) on payload upload and fetch, workflow start and +worker task completion. Other storage failures stay `external_payload_unavailable` +(503, retryable). The storage diagnostic names the failed step. The transport +manifest lists the new outcome (#589, #1233). + +## Identifier bounds + +Worker commands' `exception_class`, `exception_type` and fail-workflow +`exception.class` are capped at 255 characters, and signal, update and query +names at 255 bytes, matching their columns. An over-long value is refused with +`422 validation_failed` naming the field before anything is written; previously +it was a database error on MySQL and PostgreSQL or stored silently on SQLite. +`limits.max_operation_name_length` advertises 255 (#607, #1229). + +## Unchanged + +Worker protocol 1.20, control plane version 2, the 3.12.0 schema, retained-payload +bindings and the workflow version usage query are unchanged. The PHP Server +remains 2.11.8.