@forzalabs/remora 2.0.0 → 2.0.1

package/CHANGELOG.md

@@ -6,6 +6,18 @@ The format is based on Keep a Changelog, and this project adheres to Semantic Ve
 
 ## Unreleased
 
+## V 2.0.1 - 2026-09-01
+
+### Changed
+
+- A configuration that fails its JSON schema now reports **the mistake** rather than every reason it wasn't something else. A `oneOf` — a consumer `transform`, one alternative per transformation — makes Ajv explain why all 27 alternatives were rejected, so a `"cast": "date"` typo came out as 82 complaints whose first three were `must be equal to one of the allowed values; must have required property 'multiply'; must NOT have additional properties`, naming a transformation the field never mentioned and burying the one line that mattered. The alternatives' rejections are now collapsed: what is reported is whatever an alternative said about a value *inside* the branch — the one that got far enough in to find the real mistake — and when none did, the branch itself, listing what was actually allowed there and what the field has instead. Messages also carry the offending **value** (`must be one of: string, number, datetime, boolean (got "date")`), the **file** the resource was read from, since two resources can carry the same `name` and the one to go and edit is a path, and the **name** of the array item the pointer lands in, so `/fields/2` reads as `/fields/2 (field "created_at")` instead of sending you to count array elements
+- `remora compile` no longer prints a configuration error twice, once under `❌ Unexpected error during validation`. A resource that fails its schema is an expected outcome of `compile`, and is now reported in the same list as every other validation error
+
+### Fixed
+
+- A debug-mode run (`REMORA_DEBUG_MODE=true`) could fail every chunk with `write after end`, reported as `Affirm failed: The worker pool ran 7 thread(s) at once but was sized 2`. The executor worker **closed** the file log sink at the end of every task, but a thread outlives its task under the run-long pool, and a closed winston logger still accepts lines and writes them into an ended stream — `ERR_STREAM_WRITE_AFTER_END`, thrown from a stream nothing listens to, which inside a worker thread kills the thread and fails the chunk it was running. The sink is now flushed rather than closed; closing it resets it, so a line logged after a close re-opens a fresh sink instead of ending the process; and pointing the sink at a file it already writes no longer builds a second transport for it, which a reused thread was doing once per task. A failing sink can no longer take the process with it either — winston re-emits transport failures, and an unhandled `error` on a stream is an uncaughtException
+- A run that lost a worker thread now reports the **chunk error that killed it** instead of `The worker pool ran N thread(s) at once but was sized M`. The pool never oversubscribed: the thread tally counted spawns as concurrency, and workerpool drops a dead worker from its list synchronously while only firing its termination callback from a handler an already-exited thread never triggers — so every crashed-and-replaced thread inflated the count by one and the assertion fired ahead of the real error. Concurrency is now measured on the pool itself, and a run with dead threads says how many chunks failed and how many threads were spawned to serve them
+
 ## V 2.0.0 - 2026-09-01
 
 **Upgrading from 1.x.** Four changes can alter what an existing project does. JSON schema validation now actually runs at load, so a configuration that was silently schema-invalid fails at `compile` — run `remora compile` (or `remora graph`, which lists every problem at once without failing) before upgrading. `filters[].sql` and the `"sql"` filter operator are gone; rewrite them as `rule` filters. Filters on `number`, `datetime` and `boolean` fields now match, so a consumer whose filters were silently matching nothing will start returning rows. And `remora run -l` is now the whole run's limit rather than each worker's, so a limited run returns the number of rows asked for instead of that number multiplied by the chunk count. The `mock` command is removed — use `synth`.