9 Commits
v7 ... v12
Author SHA1 Message Date
tth 7c8359099c Merge pull request 'portainer-deploy: wait for the Swarm rollout before the smoke check (v12)' (#7) from agent/dart/portainer-wait-rollout into main
tag / tag (push) Skipped
2026-11-03 01:55:02 +01:00
tthandClaude Opus 5.5 94d0ba6b78 portainer-deploy: wait for the Swarm rollout before the smoke check (v12)
pr-check / docker (pull_request) Successful in 10s
pr-check / binary (pull_request) Successful in 1m14s
The smoke URL answered from the old task right after the stack PUT, so CI
went green before anything was deployed (birdnet-go 2026-11-03: green, then
1 min of 503) and a paused update stayed unnoticed (revaj-crm-stage
2026-11-02, new image exited 1 twice, CI green). Now the step waits until
every service whose UpdateStatus.StartedAt changed is completed and fails
on paused / rollback_*; untouched services are ignored.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 01:40:24 +01:00
renovate-bot 52fab24049 Merge pull request 'Update Node.js to e4db4c9' (#6) from renovate/ci-actions into main
tag / tag (push) Successful in 5s
2026-10-29 01:52:40 +01:00
renovate-bot 1d363f0eac Update Node.js to e4db4c9
pr-check / docker (pull_request) Successful in 11s
pr-check / binary (pull_request) Successful in 58s
2026-10-29 00:51:27 +00:00
renovate-bot dfe768f871 Merge pull request 'Update Node.js to 9481667' (#5) from renovate/ci-actions into main
tag / tag (push) Successful in 4s
2026-10-28 07:57:13 +01:00
renovate-bot fc400e752d Update Node.js to 9481667
pr-check / docker (pull_request) Successful in 9s
pr-check / binary (pull_request) Successful in 54s
2026-10-28 06:56:14 +00:00
renovate-bot 78d295d616 Merge pull request 'Update Node.js to v26' (#4) from renovate/major-ci-actions into main
tag / tag (push) Successful in 5s
2026-10-28 01:59:12 +01:00
renovate-bot 55b6b50672 Update Node.js to v26
pr-check / docker (pull_request) Successful in 11s
pr-check / binary (pull_request) Successful in 1m0s
2026-10-28 00:58:02 +00:00
tthandClaude Opus 5.5 e82c65a3c4 dtrack-sbom: never fail the job when Dependency-Track is unreachable (v8)
tag / tag (push) Skipped
The runner starts composite bash steps with -e, while the script handles
every failure itself and only warns. With dtrack.grg down, curl wrote no
resp.json, the unguarded cat of it failed and -e ended the step with
exit 1 before warn() ran - a failed build (meshcore-open run 29) instead
of a warning. set +e restores the behaviour the script was written for.

Checked with the step run under bash -e -o pipefail against an
unreachable URL (was exit 1, now ::warning:: HTTP 000 and exit 0), a
server answering 401 (warning, exit 0) and 200 (uploaded, token shown).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-26 18:26:35 +01:00
4 changed files with 90 additions and 5 deletions

No files matched your search

+1 -1
View File
@@ -38,7 +38,7 @@ jobs:
runs-on: ubuntu-latest
# No Docker CLI here, so trivy-gate falls back to the pinned binary.
container:
image: node:24-trixie-slim@sha256:43b9ba1142b8dcf33fa8f922ca1be85714adc444e0586c7cf7726daaeabc1bea
image: node:26-trixie-slim@sha256:e4db4c914bf8ad241bd24804fc25d272d89bad97e68cdc8a1414862790a02c39
steps:
- name: Tools the actions expect
run: apt-get update -qq && apt-get install -y -qq --no-install-recommends ca-certificates curl git jq >/dev/null
+12 -1
View File
@@ -104,6 +104,12 @@ binary is verified by SHA-256.
`v1` image gate + SBOM; `v2` fs scans, binary runtime, misconfig, gitleaks;
`v3` gitleaks fetches the parent of a single commit in a depth-1 checkout
(dispatch/cron runs scanned the whole tree as a root commit before).
`v8` dtrack-sbom really never fails the job: the runner's `bash -e` turned
an unreachable Dependency-Track into a failed build instead of a warning.
`v12` portainer-deploy waits for the Swarm rollout before the smoke check
and fails on a paused or rolled-back update: the smoke URL used to answer
from the old task, so a new image that never started still gave a green
deploy.
Pinned tools (Trivy binary + image, gitleaks) are bumped by Renovate
(tth/renovate). Its post-upgrade task `.gitea/update-checksums.sh`
@@ -120,7 +126,11 @@ last step of the build → `trivy-gate` → move `latest` → deploy pipeline. T
compose file in git is the single source of truth: Portainer's stored
compose is replaced on every deploy. The stack's env vars (where the secrets
live) are kept and never logged; only `vars` are overwritten. Images are
re-pulled (`pullImage`).
re-pulled (`pullImage`). The step then waits until every service the deploy
changed has finished its Swarm update (`wait-rollout`, `rollout-timeout`,
default 600 s incl. image pulls) and fails when the update is paused or
rolled back; only then is `smoke-url` checked. Services the deploy did not
touch are ignored, even when an older update of theirs is still paused.
```yaml
- uses: https://git.rfa.cz/tth-pub/ci-actions/portainer-deploy@v5
@@ -134,6 +144,7 @@ re-pulled (`pullImage`).
smoke-url: https://app.example.org/health # optional
smoke-status: "200"
smoke-timeout: "180"
rollout-timeout: "600" # optional; wait-rollout: "false" = old behaviour
```
Config files from the repo can ride along as Swarm configs. Each becomes a
+5 -1
View File
@@ -49,7 +49,11 @@ runs:
MIN: ${{ inputs.min-components }}
TRIVY: ${{ inputs.trivy-image }}
run: |
set -uo pipefail
# The runner starts composite bash steps with -e; this script handles
# every failure itself (warn, exit 0), so -e only turns an unguarded
# command into a failed job - e.g. the cat of a resp.json curl never
# wrote when Dependency-Track was unreachable.
set +e -uo pipefail
warn() { echo "::warning::SBOM $PROJECT/$VERSION: $*"; exit 0; }
[ -n "$REPORT" ] && [ -f "$REPORT" ] || warn "no report (gate skipped?)"
[ -n "$DT_KEY" ] || warn "no API key"
+72 -2
View File
@@ -5,7 +5,9 @@ description: >
replaced on every deploy, the stack's env vars (where the secrets live) are
kept and only the given variables are overwritten. Images are re-pulled.
Config files from the repo can be shipped as content-addressed Swarm
configs. Optionally waits until a URL answers with the expected status.
configs. Waits until Swarm has finished rolling out the services the
deploy changed (a paused or rolled-back update fails the step), then
optionally until a URL answers with the expected status.
inputs:
url:
@@ -46,8 +48,19 @@ inputs:
prune:
description: Remove services that are no longer in the compose file.
default: "false"
wait-rollout:
description: >
Wait until every service this deploy changed has finished its Swarm
update; fail when the update is paused or rolled back. "false" = return
right after the stack update (the old behaviour).
default: "true"
rollout-timeout:
description: Seconds to wait for the rollout (image pulls included).
default: "600"
smoke-url:
description: URL polled after the deploy until it answers smoke-status. Empty = no check.
description: >
URL polled after the rollout until it answers smoke-status. Empty = no
check.
default: ""
smoke-status:
description: Expected HTTP status of smoke-url.
@@ -70,6 +83,8 @@ runs:
CONFIGS: ${{ inputs.configs }}
CONFIG_PREFIX: ${{ inputs.config-prefix }}
PRUNE: ${{ inputs.prune }}
WAIT_ROLLOUT: ${{ inputs.wait-rollout }}
ROLLOUT_TIMEOUT: ${{ inputs.rollout-timeout }}
SMOKE_URL: ${{ inputs.smoke-url }}
SMOKE_STATUS: ${{ inputs.smoke-status }}
SMOKE_TIMEOUT: ${{ inputs.smoke-timeout }}
@@ -145,6 +160,24 @@ runs:
sets[var] = name
_, stack = api("GET", f"/api/stacks/{stack_id}")
# Rollout: remember each service's last update start before the
# PUT; afterwards only the services whose update start changed are
# this deploy's. Comparing with the previous value, not with the
# clock, keeps runner/Swarm clock skew out of it. Without this the
# smoke check below answered from the OLD task (birdnet-go
# 2026-11-03: green CI, then 1 min of 503) and a paused update
# went unnoticed (revaj-crm-stage 2026-11-02).
wait = E.get("WAIT_ROLLOUT", "true").lower() == "true"
sflt = urllib.parse.quote(json.dumps(
{"label": [f"com.docker.stack.namespace={stack.get('Name', '')}"]}))
def services():
_, out = api("GET", f"{docker}/services?filters={sflt}")
return {s["ID"]: s for s in out or []}
before = {i: (s.get("UpdateStatus") or {}).get("StartedAt")
for i, s in services().items()} if wait else {}
env = [e for e in (stack.get("Env") or []) if e.get("name") not in sets]
kept = len(env)
env += [{"name": k, "value": v} for k, v in sets.items()]
@@ -159,6 +192,43 @@ runs:
shown = ", ".join(f"{k}={v}" for k, v in sets.items()) or "nothing"
print(f"stack {stack.get('Name', stack_id)} updated from {E['COMPOSE_FILE']}: {shown}; {kept} other variable(s) kept")
if wait:
rollout_timeout = float(E.get("ROLLOUT_TIMEOUT") or 600)
start = time.time()
deadline, grace, shown_at = start + rollout_timeout, start + 20, 0.0
while True:
pending, failed, done = [], [], []
for i, s in services().items():
u = s.get("UpdateStatus") or {}
if not u or (i in before and u.get("StartedAt") == before[i]):
continue # untouched by this deploy, or newly created
name, state = s["Spec"]["Name"], u.get("State", "")
if state == "completed":
done.append(name)
elif state == "updating":
pending.append(name)
else: # paused, rollback_started / _paused / _completed
failed.append(f"{name}: {state} ({u.get('Message', '')})")
if failed:
raise RuntimeError("rollout failed - " + "; ".join(failed))
if pending:
if time.time() >= deadline:
raise RuntimeError(f"rollout not finished after {rollout_timeout:.0f} s: {', '.join(pending)}")
if time.time() - shown_at >= 30:
print(f"rollout: waiting for {', '.join(pending)}", flush=True)
shown_at = time.time()
elif done:
print(f"rollout: completed in {time.time() - start:.0f} s: {', '.join(done)}")
break
elif time.time() >= grace:
# The PUT can return before Swarm registers the update
# (birdnet-go: PUT logged 00:29:25.908, update started
# 25.929), hence the grace. Nothing after it = same
# image digest and configs, Swarm had nothing to update.
print("rollout: no service was updated by this deploy")
break
time.sleep(5)
# Older versions go once nothing mounts them; Docker refuses to
# delete a config in use (the previous task may still be draining),
# the next deploy retries.