Adds an optional priority class for run pods.
```
KUBERNETES_RUN_POD_PRIORITY_CLASS_NAME
```
When set, the value is applied as `priorityClassName` on the run pod
spec. When unset, pods are created exactly as before.
Off by default, and inert unless set. It sits beside the existing
`KUBERNETES_SCHEDULER_NAME` option and follows the same conditional
shape:
```ts
...(env.KUBERNETES_RUN_POD_PRIORITY_CLASS_NAME
? { priorityClassName: env.KUBERNETES_RUN_POD_PRIORITY_CLASS_NAME }
: {}),
```
## Verification
`typecheck --filter supervisor`, `format` and `lint` clean. No changeset
or `.server-changes/` note: off by default, no user-visible behaviour
change.
Adds KUBERNETES_RUNNER_SECURITY_CONTEXT (off | baseline | restricted), selecting how constrained the run container is.
baseline drops the capability bounding set and blocks privilege escalation. restricted additionally pins the container to a non-root uid, chosen by runtime so bun images get their own.
Default is off, so this is inert on merge.
Adds two optional env vars that rewrite the registry host of run pod images at pod creation, so a supervisor can pull from a registry in its own region. Off by default and inert unless both are set. Exact host-prefix matching, so look-alike hosts pass through untouched.
Replaces the hardcoded runner seccomp profile path with KUBERNETES_RUNNER_SECCOMP_PROFILE_PATH, and the node-24-only condition with KUBERNETES_RUNNER_SECCOMP_PROFILE_RUNTIMES (none | node-24-plus | all).
Both defaults reproduce current behaviour, so this is inert on merge. Widening the scope or turning attachment off becomes a config change rather than a deploy.
The supervisor now supports routing an organization's runs to specific
nodes. `KUBERNETES_ORG_PLACEMENT_OVERRIDES` takes JSON keyed by the
internal org ID, adding node selector entries and tolerations to that
org's run pods, e.g. to route an org onto a dedicated, tainted node
pool:
```json
{"<orgId>": {"nodeSelector": {"pool": "dedicated"}, "tolerations": "dedicated=runs:NoSchedule"}}
```
The node selector merges over the defaults (the override wins on key
collision, with a warning logged). Tolerations append to the existing
runner and scheduled-run sets. Overrides are validated at startup
similar to `KUBERNETES_RUNNER_TOLERATIONS`.
Exposed in the Helm chart as
`supervisor.config.kubernetes.orgPlacementOverrides`, where tolerations
can also be given as a list.
## Summary
`socket.io` was pinned at exactly `4.7.4` in three manifests
(`apps/webapp`, `apps/supervisor`, `packages/core`). That pin capped
`engine.io` at 6.5.4, because 4.7.4 declares `engine.io: ~6.5.2`.
Moving all three pins to `4.8.3` lifts that cap: 4.8.3 declares
`engine.io: ~6.6.0`. The webapp's direct `engine.io` devDependency moves
from `^6.5.4` to `^6.6.7` to match.
These are direct dependencies, so they are bumped in place rather than
forced with an override.
## Result
The tree previously carried two `engine.io` copies. It now carries one:
```
engine.io@6.6.8
└─┬ socket.io@4.8.3
├── @trigger.dev/core (dependencies)
├─┬ react-email
│ └── emails (devDependencies)
├── supervisor (dependencies)
└── webapp (dependencies)
```
`react-email` was already resolving `socket.io@4.8.3` in this same tree,
so that combination was already running here before this change.
## Servers move, clients do not
This bumps `socket.io` (the server) only. `socket.io-client` stays at
`4.7.5` in `packages/core` and `packages/cli-v3`, deliberately: the fix
is server-side, and clients ship inside user deployments, so leaving
them alone keeps the blast radius small. That means a 4.8.3 server will
be talking to 4.7.5 clients indefinitely, which is worth being explicit
about.
That pairing is safe because neither wire protocol changed. Both
versions report the same protocol numbers:
| | 4.7.4 | 4.8.3 |
| --- | --- | --- |
| Socket.IO protocol (`socket.io-parser`) | 5 | 5 |
| Engine.IO protocol (`engine.io-parser`) | 4 | 4 |
The version bump moves `socket.io-parser` 4.2.6 to 4.2.7 and `engine.io`
6.5.4 to 6.6.8, but the protocol constants each exports are unchanged.
The 4.8.0 changes are additive on the client (custom transport
implementations, a `tryAllTransports` option) and bug fixes on the
server.
Verified rather than assumed, with a cross-version matrix covering both
transports and both directions:
```
PASS server 4.8.3 <- client 4.7.5 websocket / polling
PASS server 4.8.3 <- client 4.8.3 websocket / polling
PASS server 4.7.4 <- client 4.7.5 websocket / polling
PASS server 4.7.4 <- client 4.8.3 websocket / polling
```
Each case exercised connect, a server-initiated emit, `emitWithAck`,
room join, room broadcast, and a binary payload. Compatibility holds in
both directions, so there is no upgrade-ordering requirement between
server and client.
`pnpm run typecheck` passes across all 57 workspaces.
Stacked on #4634.
A run controller must call the continue route to resume, so the
supervisor already knows synchronously that any checkpoint still running
for that run is pointless. It only acted on that for the compute path.
The continue route now cancels it for the Kubernetes path too, matching
what completion already does since #4493. Called after the reply so the
runner is never delayed, and skipped when there is no checkpoint client
or when the compute path owns the run. The request is bounded by a 5s
timeout so a hung call cannot leave the handler pending.
`checkpoint_cancel_requests_total{result}` records the outcome, using
the same label names as the delete path where they overlap: `sent`,
`no_client`, `not_applicable`, `http_error`.
No changeset: `CheckpointClient` is a server-only internal API, same as
#4493.
refs TRI-12915
When a run reaches a terminal state, ask the checkpoint service to
reclaim the storage its checkpoints occupied. Storage for finished runs
is not otherwise reclaimed, so nothing frees it today.
**Off by default** behind `DELETE_CHECKPOINTS_ON_COMPLETION`, and the
service-side handler ships separately, so merging this changes no
behaviour.
## Where the tenancy comes from
Addressing a run's checkpoints needs org, project, environment,
deployment version and run id. All five are already in hand at
`attempt.complete`, and three are **signed** by the deployment token:
| Value | Source | Trust |
| -- | -- | -- |
| org | claim `org_id` | signed |
| environment | claim `environment_id` | signed |
| deployment version | claim `deployment_version` | signed |
| project ref | `x-trigger-workload-project-ref` header |
runner-supplied |
| run | route param | runner-supplied |
`authorizeWorkloadRequest` previously returned only `environment_id`,
and only in enforce mode, so it now also returns the verified `claims`.
That difference is deliberate and documented on the method: claims are
used to address a run's **own** resources locally, never to scope the
platform, which is why `environmentId` stays enforce-only.
The two runner-supplied values are safe because the signed ones are
outermost - a runner lying about either can only name something inside
its own org and environment, and a project ref that doesn't pair with
its signed environment matches nothing. The run id is read from
`params.runFriendlyId`, the same value the platform just validated,
rather than from the body or a header. Where both a claim and a header
exist (`deployment_version`), the claim wins.
## Placement
The call sits after `reply.json(...)`, so the runner sees no added
latency - the same shape the suspend route already uses. The service
enqueues and returns 202, so it is one fast local hop.
Terminal means `RUN_FINISHED` **or `RUN_PENDING_CANCEL`** - a run
cancelled mid-execution never restores, and skipping it would leave its
storage behind. Retries are excluded deliberately: reclamation is
per-run, so a retry is covered by the final completion.
Also gated on `!snapshotService`, so it stays inert where checkpoints
aren't the kind this reclaims.
## Observability
`checkpoint_delete_requests_total{result}` counts `sent` **and every
reason we decide not to send**: `disabled`, `not_terminal`, `no_claims`,
`no_project_ref`, `http_error`.
The negative labels are the point - without them, "no requests are
happening" looks identical to the feature being switched off.
`no_claims` is reachable even under enforcement, since enforce only
rejects a *present-but-invalid* token; an absent or legacy id still
passes with no claims attached.
## Notes for review
- **No changeset**: `CheckpointClient` is `core/v3/serverOnly`, an
internal service-to-service API rather than customer-facing surface.
- **No `.server-changes/` note**: there is nothing a dashboard user
would notice here. Happy to add one if you disagree.
- `pnpm run typecheck` can't complete in my checkout -
`@trigger.dev/database` fails to build on a missing `tsc` in the pnpm
store, unrelated to this diff. Verified with `tsc --noEmit` against the
supervisor project instead: **zero errors in `apps/supervisor/src`**.
Worth noting it caught a real bug here - the completion response is
wrapped, so the status is `data.result.attemptStatus`.
refs TRI-12789
The dequeue brake released the moment its signal became unreadable.
`refresh()` caught any error from `source.read()` and set the verdict to
`null`, which `computeEngaged()` treats as not-engaged — so a few failed
reads dropped an engaged brake, silently, with no log and no metric.
That handling was symmetric while the risk is not. A source that has
stopped answering correlates with the pressure the brake exists for, so
releasing on read failure gives up protection at exactly the wrong
moment; holding too long only costs throughput.
Now a failed read keeps the last verdict instead of discarding it. The
verdict then ages normally, so the existing `maxVerdictAgeMs` check
becomes the grace window and still bounds how long a dead source can
hold the brake — a permanently unreachable source releases it rather
than pinning dequeuing forever. Because `computeEngaged()` only consults
staleness for an *engaged* verdict, a released one is unaffected and
stays released.
The default grace moves from 15s to 120s, comparable to how long the
brake normally stays engaged.
One guard worth calling out: holding is only safe when something bounds
it, so when `maxVerdictAgeMs` is unset the previous discard behaviour is
kept. Otherwise an unbounded hold could pin the brake indefinitely.
Read failures were previously invisible — the catch block neither logged
nor counted. Adds a `read_failures_total` counter, plus an error log on
the transition into failure rather than once per tick, since the refresh
loop runs every second.
The post-release ramp needs no change: it anchors off the
engaged-to-released transition, so a grace-window release still ramps
back up instead of snapping to full rate, which is what you want after a
blind period.
Tests cover holding while reads fail, releasing past the max age, and
the existing unbounded-config paths are unchanged.
## Summary
Self-hosted Kubernetes deployments can now add tolerations to run pods,
so runs
can schedule onto tainted nodes. Previously the only way to do this was
to patch
the supervisor.
`KUBERNETES_RUNNER_TOLERATIONS` takes a comma separated list of
`key=value:effect`, or `key:effect` to tolerate any value. It applies to
every
run pod, and for runs from a schedule tree it merges with the existing
`KUBERNETES_SCHEDULED_RUN_TOLERATIONS`. Left unset, nothing changes: no
tolerations are added and the pod spec leaves the field off entirely.
The Helm chart takes it as a list:
```yaml
supervisor:
config:
kubernetes:
runnerTolerations:
- dedicated=runs:NoSchedule
- spot:NoExecute
```
## Naming
The issue proposed `KUBERNETES_WORKER_TOLERATIONS`. This ships as
`KUBERNETES_RUNNER_TOLERATIONS` instead, because `RUNNER_*` is already
the prefix
for run pod settings (`RUNNER_HEARTBEAT_INTERVAL_SECONDS`,
`RUNNER_ADDITIONAL_ENV_VARS`, and `DOCKER_RUNNER_NETWORKS` for the
Docker
equivalent), whereas "worker" refers to the supervisor itself throughout
this app.
## Validation
Keys and values are checked against the Kubernetes naming rules when the
supervisor starts, so `dedicated=prod runs:NoSchedule` fails immediately
with a
message naming the offending entry. Without that check a bad value is
accepted at
startup and then rejected by the API server on every pod create, which
stops all
runs with the cause buried in an API error.
`KUBERNETES_WORKER_NODETYPE_LABEL` is
trimmed and validated for the same reason: surrounding whitespace is not
valid in
a label value, so a padded value fails every pod create today.
## Node selector off switch
`KUBERNETES_WORKER_NODETYPE_LABEL` accepts an empty string to skip the
node
selector entirely, so runs schedule on any node. This already worked and
the Helm
chart has always shipped it empty, but it was not documented. It is now.
The issue also asked for general node affinity configuration. That is
not
included: the node selector off switch plus tolerations covers the
reported
problem, and a free form affinity setting is a much larger config
surface to
commit to.
Fixes#4458
The pod-count backpressure source read
`apiserver_storage_objects{resource="pods"}` from an apiserver
`/metrics` scrape. That gauge is a periodically-refreshed cached count,
and it is served by whichever apiserver replica the scrape lands on —
replicas disagree with each other at the same instant, by enough to
swamp the engage/release hysteresis band. Engage and release timing was
therefore partly a function of scrape routing.
This replaces it with a single `limit=1` list of the workload namespace
and computes `remainingItemCount + items.length`. One pod object
transferred, no informer, no watch cache.
Two request-shape constraints are load-bearing and called out in the
code: passing a label or field selector makes the apiserver omit
`remainingItemCount` entirely, and setting `resourceVersion` serves a
cached count rather than a quorum read. Neither is passed.
`remainingItemCount` is only set when the list is truncated, so
`_continue` is the truncation signal — if it is absent the returned page
is the whole collection and `items.length` is already exact. If the list
*is* truncated and the count is missing or implausible, the fetcher
throws rather than guessing.
Failure semantics are unchanged: a throw lands in the monitor's existing
catch, exactly as the previous parse did. The hysteresis, verdict shape,
and gauge are untouched. RBAC is unchanged — the existing role already
grants `pods: list`.
The `/metrics` non-resource grant in the deployment role becomes unused,
and the scrape-timeout env var is now a slight misnomer. Both left alone
deliberately: the grant may be wanted again for other apiserver signals,
and renaming the var would need a coordinated config change for no
behavioural gain.
Tests cover the not-truncated, truncated, missing-count, negative-count
and timeout paths.
Adds an optional `TRIGGER_WARM_START_DISPATCH_URL`. The warm-start
dispatch request uses it when set, otherwise falls back to
`TRIGGER_WARM_START_URL`, so the dispatch target can differ from the
default warm-start URL. No behavior change when unset.
Adds Prometheus metrics so the supervisor's outbound HTTP calls are
observable - including client-side failures that previously only
surfaced as a log line.
- `supervisor_outbound_request_total{name, method, status, outcome}` -
counts every outbound request. `outcome` separates a transport failure
(`network_error`), an HTTP error response (`http_error`), a response
that failed schema validation (`invalid_response`), and success (`ok`).
- `supervisor_outbound_request_duration_seconds{name, outcome}` -
latency histogram. Leaner labels than the counter (no `status`) to avoid
bucket×label cardinality; buckets match the existing dequeue-latency
histogram since these calls share the same retrying HTTP client and
long-poll envelope.
Coverage:
- The warm-start request (a one-off `fetch`) - instrumented inline; the
response status code is now also included in the failure log (it was
previously dropped).
- All worker API client calls (`SupervisorHttpClient`: dequeue, run
attempt start/complete, heartbeats, snapshots, continue, suspend,
debug-log, connect) - routed through a single instrumented `request()`
helper that reports via an optional `onHttpRequestComplete` callback on
the client, which the supervisor wires into the counter + histogram.
Low cardinality by design: `name` is a **static per-endpoint label**
(e.g. `dequeue`, `start_run_attempt`), never the interpolated URL - so
no run/snapshot IDs land in labels, mirroring the templated `route`
labels on the inbound HTTP server.
Registered on the existing metrics registry, exposed on `/metrics` with
no new wiring. Internal-only change (no package release needed), so the
changelog note is a single `.server-changes` entry.
Add a Prometheus gauge `workload_token_enforcement_mode` set to 1 for
the active `WORKLOAD_TOKEN_ENFORCEMENT` value
(`disabled`/`log`/`enforce`), emitted at startup on the shared registry.
The existing mint/verify counters don't distinguish `log` from `enforce`
(the verify outcome is recorded before the reject decision), so
dashboards can't tell which mode a cluster is running. This gauge makes
the active mode queryable at a glance. Supervisor typecheck passes.
## Summary
Adds experimental Node.js 24 and 26 task runtimes through the
`experimental-node-24` and `experimental-node-26` config values.
Existing runtime defaults and the `node`, `node-22`, and `bun` behavior
remain unchanged. The unprefixed `node-24` and `node-26` config values
remain unavailable until the runtimes are ready for general use.
## Design
Experimental config values normalize to canonical runtime identifiers
before build manifests are created, keeping deployment metadata and
execution behavior consistent. Kubernetes task pods also use the
runtime-default seccomp profile so modern Node.js versions fall back
from io_uring to checkpoint-compatible system calls.
## Summary
Updates the internal development, CI, and runtime-image Node version to
24.18.0. SDK compatibility coverage continues to include Node 20, 22,
24, and 26.
The Node type definitions and the package-manager lockfiles now resolve
against Node 24 types.
## Summary
v3 (the engine that ran the SDK v3 era, internally
`RunEngineVersion.V1`) is end-of-life. Following the removal of the v3
execution apps
([#4194](https://github.com/triggerdotdev/trigger.dev/pull/4194)) and
the legacy dev websocket
([#4198](https://github.com/triggerdotdev/trigger.dev/pull/4198)), this
removes the remaining v3 execution stack from the server.
Clients still on v3 (an old SDK or CLI that has not upgraded) keep
getting a clear "upgrade to v4" response. Triggers, batch triggers,
reschedules, and deploys that resolve to v3 are rejected with a graceful
4xx pointing at the migration guide, never a 5xx, so a stale client
cannot affect server health. Self-hosted instances still running v3
should stay on the 4.5.x release line until they migrate.
## What is removed
- The MarQS queue and its shared/dev queue consumers.
- The v3 socket.io namespaces (coordinator, provider, shared-queue) and
the v3 run lifecycle services (attempt, checkpoint, and batch-resume).
- The graphile-worker background job system; all live jobs already run
on `@trigger.dev/redis-worker`.
- The `DEPRECATE_V3_ENABLED` flag: v3 is now rejected unconditionally,
so the flag is gone.
- Unused v3 exports from `@trigger.dev/core` (the `v3/zodNamespace`
subpath and the legacy socket message catalogs) and the now-dead MarQS
environment variables.
## What stays
The v4 engine is untouched. The graceful v3 rejection boundary stays,
`determineEngineVersion` still detects a v3 project so it can reject it,
and the batch service plus batch-completion worker stay for current
clients. Live queue concurrency limits and metrics now read from the v4
run engine instead of MarQS, and a brand-new dev environment now
defaults to v4.
## Dependency cleanup
Removes webapp dependencies left unused by this change: `seedrandom` and
`semver` (only the removed v3 code used them) plus a set that was
already dead, their orphaned `@types` packages, and two dead files. Adds
a `knip:deps` script and a `knip.json` config so unused dependencies can
be found the same way going forward.
## What
Adds the missing `COPY scripts/retry-prisma-generate.mjs` to the
supervisor `Containerfile` builder stage, before `RUN pnpm run
generate`.
## Why
The `generate` scripts in `internal-packages/database` and
`internal-packages/run-ops-database` shell out to
`scripts/retry-prisma-generate.mjs`. The supervisor build never copied
that file into the image, so `pnpm run generate` failed:
```
@internal/run-ops-database:generate: Error: Cannot find module '/app/scripts/retry-prisma-generate.mjs'
```
This is the same failure class as #4156 (webapp Dockerfile). The
supervisor `Containerfile` is the **only other** build file that runs
`pnpm run generate` — the coordinator / docker-provider /
kubernetes-provider Containerfiles don't, so this completes the fix.
## Verification
Local `docker build` of the supervisor `Containerfile` builder target —
result appended below once the build completes.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Stream dev logs over a local telnet/TCP socket. `trigger dev` mirrors
its terminal output on port 6767 by default (override with
--telnet-logs-port or TRIGGER_DEV_TELNET_LOGS_PORT, 0 disables). webapp,
supervisor, and coordinator each expose an opt-in stream gated on a
per-service *_TELNET_LOGS_PORT env var. New
@trigger.dev/core/v3/telnetLogServer module (localhost-only,
backpressure-safe, plain-text) plus optional static Logger.onLog /
SimpleStructuredLogger.onLog sinks.
Then you (or your agent) can use `nc` to connect and filter out the
stream.
<img width="1103" height="239" alt="image"
src="https://github.com/user-attachments/assets/b4d47efc-8a57-4185-a159-10f2806627ae"
/>
Bumps the internal/toolchain Node version to the latest 22.x LTS
(`22.23.1`) and standardises it across the repo. Scope is the **platform
toolchain + the repo's own runtime images** (all `20 → 22` *upgrades*,
off the now-EOL node 20).
### Main changes
- Node `20.20.2 → 22.23.1` across all CI workflows, `.nvmrc`,
`CONTRIBUTING.md`, and the OSS `docker/Dockerfile` (digest-pinned).
- `@types/node → 22.20.0` (root dep + pnpm `overrides`, so the whole
workspace resolves to it); lockfile regenerated.
- `sdk-compat` matrix: adds Node 24 + 26 (keeps 20, still in `engines`).
- **App runtime images → node 22** (were on EOL node 20):
`apps/coordinator` → `node:22.23.1-bookworm-slim`;
`apps/docker-provider` + `apps/kubernetes-provider` → `node:22-alpine`
(reusing the exact digest `apps/supervisor` already runs, so all four
worker images are now identical). Stage aliases renamed off `node-20`.
### Possible issues / test notes
- `@types/node` 22.x can surface new TS errors — typecheck (now on 22)
is the gate.
- **Smoke-test the v3 worker path** — `coordinator` (`crictl`/CRI calls)
and the docker/kubernetes providers (talking to their daemons) now run
on node 22 (alpine/musl for the providers). Upgrade off EOL so low-risk,
but it's deployed runtime code with its own `publish-worker.yml`
pipeline.
Once this is merged, oxlint is at a pretty sensible baseline.
**Enable `no-unused-vars`, `typescript/consistent-type-imports`, and
`import/no-duplicates` lint rules**
Turns on three previously-disabled oxlint rules across the monorepo and
fixes all violations:
- **`no-unused-vars`** – enabled as an error with standard ignore
patterns: unused function arguments are ignored by default (`args:
"none"`), variables/caught errors/destructured array elements prefixed
with `_` are allowed, and rest siblings are permitted.
- **`typescript/consistent-type-imports`** – enforced as an error; all
type-only imports now use the `import type` syntax.
- **`import/no-duplicates`** – enforced as an error; duplicate import
statements from the same module have been merged.
The remaining commits clean up the violations found across the codebase:
removing unused variables/imports/type aliases, adding `_` prefixes to
intentionally unused bindings, fixing duplicate imports, and converting
value imports to `import type` where appropriate.
The supervisor image build has been failing since `@trigger.dev/core`
gained
an `ai` peer dependency. `turbo prune` (2.5.4) generates a pruned
lockfile
that references the `ai@6.0.116(zod@3.25.76)` snapshot without including
the
entry itself, which causes `pnpm fetch --frozen-lockfile` to abort.
Bumping to 2.10.0 fixes the pnpm v9 peer dep snapshot pruning. Updated
both
Containerfiles for consistency.
Example failure here:
https://github.com/triggerdotdev/trigger.dev/actions/runs/28225353375/job/83618124564
Broken since:
c06005b3
Adds an in-process backpressure signal that pauses dequeuing when the
Kubernetes cluster is saturated, so work overflows cheaply in the queue
instead of piling up as unschedulable pods. Saturation is read by
scraping the apiserver's total pod-object count
(`apiserver_storage_objects{resource="pods"}`) and applying an
engage/release threshold with hysteresis - a single lightweight
aggregate scrape, not a pod listing.
Backpressure sources are now evaluated independently and OR'd: each
source has its own enable and dry-run flag, and the supervisor engages
if any enabled source trips. This adds the pod-count source alongside
the existing one without changing it, and is extensible to more sources
later. Off by default.
The scrape uses the in-cluster kubeconfig over `https` so TLS verifies
against the cluster CA (the fetch-options helper attaches the CA as an
`https.Agent`, which the global `fetch` ignores - that path silently
dropped the CA). Enabling the pod-count source requires the supervisor's
service account to be granted `get` on the `/metrics` non-resource URL;
that RBAC and the per-deployment env wiring are operator-side and live
elsewhere.
New config (pod-count source):
`TRIGGER_DEQUEUE_BACKPRESSURE_POD_COUNT_ENABLED` (default false),
`_POD_COUNT_DRY_RUN` (default true), `_POD_COUNT_ENGAGE` /
`_POD_COUNT_RELEASE` (hysteresis thresholds), `_POD_COUNT_REFRESH_MS`
(scrape interval, default 5s). The existing source's flags are
unchanged.
Observability: a `supervisor_cluster_pod_count` gauge, and the pod-count
monitor's metrics are namespaced (`supervisor_backpressure_pod_count_*`)
so the existing backpressure metrics keep their names.
Follow-up to #3992, which gated the send runner-side - but only for new
runner images. Existing runners still POST a debug log per line.
When `SEND_RUN_DEBUG_LOGS` is off (default), the route now drops the
request immediately: `skipBodyParsing` skips the body read/parse, a bare
handler returns 204, no wide event. The route stays registered so it
avoids the `No route match` error log; the only per-request log left is
the framework's `logger.debug` trace, suppressed at the default `info`
level. Still counted by request metrics, and 204 is non-retryable so no
retry storm.
Adds a `skipBodyParsing` flag to the internal HTTP server.
Runners were POSTing a debug log to the supervisor for every log line -
one request per line, unbatched and unconditional. The supervisor
already has a `SEND_RUN_DEBUG_LOGS` toggle (off by default) that
discards them on receipt, but the runner fired the request regardless,
so the traffic hit the supervisor either way.
This gates the send at the source. The runner now reads
`TRIGGER_SEND_RUN_DEBUG_LOGS` (off by default, injected by the
supervisor from its existing `SEND_RUN_DEBUG_LOGS` setting) and skips
the POST entirely when disabled. Local log output is unchanged. Dev runs
use a separate path and are unaffected.
### Problem
Firestarter's `didWarmStart: true` means the response was written to a
long-poll socket — not that the runner received it. A silently dead
poller (no FIN, e.g. a VM torn down mid-poll) leaves the dispatched run
stuck in `PENDING_EXECUTING` until the run engine's heartbeat redrive,
and each redrive burns a queue redelivery toward
`TASK_RUN_DEQUEUED_MAX_RETRIES`.
### Change
After a warm-start hit, the supervisor retains the `DequeuedMessage`
(TimerWheel, default 10s), then probes the existing `getLatestSnapshot`
API. If the run is still on the exact dequeued snapshot, no runner ever
acted — it falls through to the regular cold-create path. Recovery: ~10s
+ cold start, no new APIs, no CLI changes.
- **Double-start safe**: `startRunAttempt` runs under a per-run lock and
409s stale snapshot ids, so a reviving runner and the fallback workload
can't both execute; the loser exits before running anything.
- **Probe errors → do nothing**: healthy runners legitimately act late
during platform brownouts (nested attempt-start retries), so falling
back on uncertainty would stampede duplicates. The heartbeat redrive
stays as the backstop (also covers supervisor restarts dropping timers).
- **Off by default**: `TRIGGER_WARM_START_VERIFY_ENABLED` (+
`TRIGGER_WARM_START_VERIFY_DELAY_MS`, 1–60s, default 10s). Disabled =
complete no-op. Works for all workload managers (compute/k8s/docker)
since it hooks the shared dequeue path.
- Emits `warmstart.verify` wide events (`outcome: delivered | fallback |
probe_error`), making the silent-loss rate directly measurable.
Adds a `workload_create_duration_seconds` Prometheus histogram to the
supervisor, observed around the workload manager `create()` call:
- `backend` label: `kubernetes` | `compute` | `docker` — set once from
the configured workload manager
- `outcome` label: `success` | `error` — the per-outcome counts double
as a create error rate
Registered on the supervisor's existing metrics registry, so it's
exposed on the existing `/metrics` endpoint with no config changes.
Notes:
- Covers cold creates only; warm starts and restores return before
reaching `create()`.
- A create may include backend-internal retries, so one observation can
span multiple attempts.
- Fixed low cardinality: 2 active label sets per deployment × 10
buckets.
The compute suspend flow delays snapshots by `snapshotDelayMs` (~30s) so
short-lived waitpoints skip the snapshot entirely, with the intent that
a run continuing before the delay expires cancels the pending snapshot.
But the only `cancel()` call site was the `/continue` action, which
runners only invoke when restoring from an already-taken snapshot — so
pending snapshots were never cancelled (zero `snapshot.canceled` events
ever emitted in prod). When a run resumed and completed inside the
window, the stale snapshot fired ~30s later anyway, pausing the VM 6–13s
mid warm-start long-poll; the frozen guest couldn't fire its abort timer
or send a FIN, causing stalls and run-engine driven retries.
### Change
- Cancel the pending snapshot on `attempt.complete` — after the platform
accepts the completion, before the HTTP reply (so it can't reorder with
the runner's next `/suspend`).
- Cancel on `runDisconnected` (crash, exit, or run replaced on the
socket).
- Both cancels are guarded by a runnerId match (new
`TimerWheel.peek()`): a stale duplicate runner for a reassigned run must
not cancel the fresh runner's pending snapshot. A missing runnerId falls
through to an unconditional cancel (the pre-existing `/continue`
behavior is unchanged).
Waitpoint suspensions keep the runner socket connected and the attempt
incomplete, so neither hook touches a snapshot that is still wanted.
Known limitation (fail-safe direction): `socket.data.runnerId` is frozen
at the websocket handshake, so after a same-supervisor restore the
disconnect-path guard refuses the cancel. The `attempt.complete` path
uses the runner's current header id and is unaffected.
`ComputeWorkloadManager.create` swallows gateway errors currently, so a
cold start that fails placement (e.g. a netns slot with a busy tap, a
full node disk) silently abandons the dequeued run until the run
engine's `PENDING_EXECUTING` heartbeat timeout redrives it via stall
detection.
### Changes
- Retry `instances.create` with short backoff (default 3 attempts, 250ms
backoff), recording `createAttempts` in the wide event.
- **Only statuses where the create definitely did not commit are
retried**: 500 (agent/fcrun create failed) and 503 (no placement).
502/504 are excluded — the gateway emits those when it fails to reach
the node or read its response, which can happen *after* the agent
committed the create; the gateway only records the instance name on a
clean 201, so a same-name retry would miss the collision check and could
double-create the VM on another node. Network-level fetch failures are
retried (if the gateway processed the create, its name index is
populated and the retry 409s harmlessly). Timeouts are not retried.
- **Retry attempts after a 5xx use a deterministic `-rN` name suffix**:
a failed create can leave its name registered until async cleanup runs.
Attempt 1 keeps the unsuffixed name.
## Summary
Adds a second backend for the realtime runs feed (`useRealtimeRun`,
`subscribeToRunsWithTag`, `subscribeToBatch`), built to stay healthy
when a single busy environment has many subscribers watching many runs
at once. It is gated behind a feature flag with the existing backend as
the default, so nothing changes for users until it is enabled per
environment.
## Design
A run change is published once, as a small self-describing record, to a
single per-environment channel. Every feed is then a predicate over that
one stream rather than owning a channel:
- A per-instance router indexes the currently-held feeds by run, tag,
and batch. When a run changes it hydrates the affected rows once and
serializes them once, then fans the result to every matching feed. One
hot shared tag watched by many subscribers costs a single database query
and serialize, not one per subscriber.
- Feeds that don't match a change are never woken, wake delivery per
environment is coalesced on a leading edge (250ms default) so a burst of
changes costs one wake, and cold reads coalesce onto a single
short-TTL-cached resolve.
- An admission gate bounds how many cold ClickHouse resolves run
concurrently, so a mass reconnect across many distinct filters queues
instead of stampeding the database.
- Changes that land while a client is between long-polls are delivered
on its next poll instead of waiting for the periodic backstop: each
environment buffers its recent change records, subscriptions linger
briefly after the last feed closes, and a newly-armed poll replays
exactly the connection's gap.
- The per-connection replay cursors behind that are shared across
instances via Redis (a single timestamp each), so a poll landing on a
different instance behind the load balancer still reads the connection's
true gap instead of falling back to a cold resolve. Cursor reads have a
bounded deadline and degrade to the cold-read path on any Redis trouble.
- Tag subscriptions with multiple tags match runs carrying all of the
tags, mirroring the existing backend's filter semantics, and live
long-polls hold for about 20 seconds to match its cadence.
- The per-environment channel supports Redis Cluster sharded pub/sub, so
the wake path scales horizontally across shards by environment.
- The backend reports its health through OpenTelemetry metrics (delivery
lag, poll resolution paths, backstop outcomes, replay and cursor-store
activity), with a provisioned Grafana dashboard for local development.
Everything is behind the feature flag and tunable via env vars; the
existing backend remains the default.
Add an optional network_labels field to the internal compute client's
create and restore request schemas and forward per-VM endpoint labels on
both paths, so a restored VM keeps the same labels as a freshly-booted
one. Mirrors the label the Kubernetes workload manager already sets on
the run pod.
---------
Co-authored-by: nicktrn <55853254+nicktrn@users.noreply.github.com>
The supervisor can now pause dequeuing - and freeze consumer-pool
scale-up - when a backpressure signal says the cluster can't place more
work, then ramp dequeuing back up gradually once it clears. The signal
is a verdict published to a Redis key by a cluster-side component; the
supervisor reads it on a short refresh and gates `preDequeue` on it.
Off by default (`TRIGGER_DEQUEUE_BACKPRESSURE_ENABLED`). Everything
fails open: a missing, stale, or unreadable verdict never pins the
brake, and the hot-path read is a synchronous cached lookup with no I/O.
The scale-up freeze leaves scale-down untouched, and on release the
resume is ramped so a deep queue isn't hammered all at once.
Dry-run is on by default (`TRIGGER_DEQUEUE_BACKPRESSURE_DRY_RUN`): even
once enabled it only logs what it would have done, and surfaces the
computed state through metrics, until explicitly set to act. Prometheus:
`supervisor_backpressure_engaged`, `_dry_run`,
`_skipped_dequeues_total`.
Refs TRI-5354
## Summary
Scheduled runs and their descendants can now be routed to a dedicated
per-region worker queue, processed by a separate worker fleet, so a
burst of scheduled crons no longer competes with standard and agent runs
for the same queue and inflates their startup latency. It is off by
default and enabled per organization via a feature flag (with a global
default), so nothing changes until it is turned on.
## Design
At trigger time, any run whose lineage originates from a schedule
(`rootTriggerSource === "schedule"`, which already propagates from a
scheduled run down to all of its children) gets its worker queue
suffixed with `:scheduled`. The worker queue name is an opaque string
persisted on the run and used verbatim by enqueue and dequeue, so this
needs no Lua, message-envelope, or concurrency changes. Concurrency
stays keyed by environment and queue, not by worker queue.
On the consumer side, the dequeue endpoint gains an optional
`queueClass` selector. A supervisor sends `queueClass: "scheduled"` and
the server derives the actual queue from the worker's own group, so a
token can only ever reach its own region's queues. A fleet picks its
class with the `TRIGGER_WORKER_QUEUE_CLASS` env var (`default` or
`scheduled`), so a dedicated scheduled fleet can run alongside the
standard one.
Verified end to end against a local managed-worker setup: scheduled runs
route to the dedicated queue, are drained only by the scheduled fleet,
and standard runs are left untouched.
Adds wide-event observability for the supervisor: one flat-keyed JSON
line per dequeue iteration, workload-server route, and run socket
lifecycle event. Events carry `trace_id` sourced from the inbound W3C
traceparent plus `meta.run_id` and related identifiers, so they join
across services by run.
The outbound warm-start POST also forwards the inbound traceparent so
the upstream receiver continues the same trace instead of minting a new
one.
Off by default behind `TRIGGER_WIDE_EVENTS_ENABLED`. With the flag off,
no events are emitted, no ALS state is allocated, and the outbound
warm-start request is unchanged — every call site was audited to confirm
the off path is byte-identical to current behavior.
Dequeue-path phase timings recorded under `phase.<name>.duration_ms`:
`restore`, `warm_start`, `workload_create`. A `path_taken` extra
distinguishes `restore` / `warm_start` / `cold_create` /
`skipped_no_image`.
Refs TRI-9480.
Workloads bundled with CLI versions before v4.4.4 use a strict zod enum
for `checkpoint.type` that only allows DOCKER and KUBERNETES. When a
customer's runs are routed via the compute path, those old runners
receive `type: "COMPUTE"` on `/snapshots/since/...` and `/dequeue`
responses and fail validation - blocking silent migration of existing
deployments.
The workload never reads the field - only validates the shape. Rewriting
COMPUTE -> KUBERNETES on the way out lets older runners keep parsing
while the database and internal services keep the real value. Limited to
the two workload-facing endpoints whose response includes a checkpoint;
`/continue`, `/attempts/start`, `/attempts/complete` all return shapes
without one.
Followup to #3114.
Adds `KUBERNETES_POD_DNS_NDOTS_OVERRIDE_ENABLED` flag (off by default)
that overrides the cluster default and sets `dnsConfig.options.ndots` on
runner pods (defaulting to 2, configurable via
`KUBERNETES_POD_DNS_NDOTS`).
Kubernetes defaults pods to `ndots: 5`, so any name with fewer than 5
dots, including typical external domains like `api.example.com`, is
first walked through every entry in the cluster search list
(`<ns>.svc.cluster.local`, `svc.cluster.local`, `cluster.local`) before
being tried as-is, turning one resolution into 4+ CoreDNS queries (×2
with A+AAAA).
Using a lower `ndots` value reduces DNS query amplification in the
`cluster.local` zone.
Adds region-level gating so MICROVM regions are only visible and usable
by orgs with the `hasComputeAccess` feature flag. Admins and explicit
allowlist behavior unchanged.
- New shared helper (`regionAccess.server.ts`) with
`resolveComputeAccess`, `defaultVisibilityFilter`, and
`isComputeRegionAccessible`
- `RegionsPresenter` filters out MICROVM regions for non-compute orgs
- `SetDefaultRegionService` blocks setting a MICROVM region as default
without compute access
- `WorkerGroupService` blocks triggering runs in MICROVM regions without
compute access
- `computeTemplateCreation` refactored to use shared
`resolveComputeAccess`
- Updated snapshot callback schema
Adds support for taint tolerations for scheduled runs. Useful for
selectively tolerating taints on dedicated node pools.
The new `KUBERNETES_SCHEDULED_RUN_TOLERATIONS` env variable accepts a
comma-separated list in the format key=value:effect (or key:effect for
the Exists operator).
Drive-by: renames all `KUBERNETES_SCHEDULE_*` affinity env vars to
KUBERNETES_SCHEDULED_RUN_* for clarity — this feature isn't used in
production yet or published in a tagged image; the name change is fine.
Scheduled runs create predictable hourly spikes that compete with
on-demand runs for node capacity. Runs triggered "on-demand" via the
SDK, API, or dashboard, are more sensitive to cold start latency since
users are typically
waiting on the result. When a burst of scheduled runs lands at the top
of the hour, it can saturate the shared pool resources causing
contention, affecting cold starts across the board.
The idea in this change is to absorb these periodic spikes in a
dedicated pool without affecting the cold starts of on-demand runs.
Scheduled runs are inherently less sensitive to cold starts.
### Changes in this PR
Follows up on run annotations (#3241), which made trigger origin
available on every run in the tree. This PR exposes
annotations at dequeue time to the supervisor. This enables scheduling
decisions based on trigger source.
The affinities are soft preferences at schedule time, so runs fall back
gracefully if the target pool is out out of capacity.
Adds optional pod affinity so pods from the same project prefer
scheduling on the same node. This can help improve image cache hit
rates; subsequent pods benefit from already-pulled image layers,
reducing startup time.
Complements the built-in ImageLocality scheduler plugin by helping
during burst scheduling scenarios. Pod affinity sees scheduled pods
immediately, while ImageLocality only sees images after they're fully
pulled.
Configuration:
- `KUBERNETES_PROJECT_AFFINITY_ENABLED` - Enable/disable (default:
false)
- `KUBERNETES_PROJECT_AFFINITY_WEIGHT` - Scheduler weight 1-100
(default: 50)
- `KUBERNETES_PROJECT_AFFINITY_TOPOLOGY_KEY` - Topology key (default:
kubernetes.io/hostname)
Uses soft (preferred) affinity so pods always schedule even if preferred
node is full.
<!-- devin-review-badge-begin -->
---
<a
href="https://app.devin.ai/review/triggerdotdev/trigger.dev/pull/2995">
<picture>
<source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-open-in-devin-review-dark.svg?v=1">
<img
src="https://static.devin.ai/assets/gh-open-in-devin-review-light.svg?v=1"
alt="Open with Devin">
</picture>
</a>
<!-- devin-review-badge-end -->
Adds support to configure CPU/memory request ratios per machine preset.
Falls back to the global request ratio configs if no specific override
is specified.
Runs across different machine presets have different usage patters, so
this enables use to manage the available capacity better.
nicktrn
