starred/orca
1
0
Fork 0
mirror of https://github.com/stablyai/orca.git synced 2026-10-08 00:02:38 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: harden mobile push delivery and deployment recovery

Browse Source
This commit is contained in:
Jinwoo-H
2026-09-06 18:42:05 -04:00
parent 41ace23a51
commit b71ffe09fe
56 changed files with 1439 additions and 308 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/cloud-push-deploy.yml
+11 -6
View File
@@ -130,11 +130,14 @@ jobs:
run: |
set -euo pipefail
tag="c${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
echo "CANDIDATE_TAG=${tag}" >> "${GITHUB_ENV}"
echo "CANDIDATE_REVISION=${SERVICE_NAME}-${tag}" >> "${GITHUB_ENV}"
gcloud run deploy "${SERVICE_NAME}" \
--project "${GCP_PROJECT_ID}" \
--region "${GCP_REGION}" \
--image "${IMAGE}" \
--tag "${tag}" \
--revision-suffix "${tag}" \
--no-traffic \
--quiet
candidate="$(gcloud run services describe "${SERVICE_NAME}" \
@@ -142,8 +145,7 @@ jobs:
| jq -er --arg tag "${tag}" \
'[.status.traffic[] | select(.tag == $tag)]
| if length == 1 then .[0] else error("tagged candidate is not unique") end')"
echo "CANDIDATE_TAG=${tag}" >> "${GITHUB_ENV}"
echo "CANDIDATE_REVISION=$(jq -r '.revisionName' <<< "${candidate}")" >> "${GITHUB_ENV}"
test "$(jq -r '.revisionName' <<< "${candidate}")" = "${SERVICE_NAME}-${tag}"
echo "CANDIDATE_URL=$(jq -r '.url' <<< "${candidate}")" >> "${GITHUB_ENV}"
# A tagged revision is directly addressable and sits outside the service-wide cap, so the
@@ -224,6 +226,7 @@ jobs:
shell: bash
run: |
set -euo pipefail
echo "TRAFFIC_SHIFT_ATTEMPTED=true" >> "${GITHUB_ENV}"
gcloud run services update-traffic "${SERVICE_NAME}" \
--project "${GCP_PROJECT_ID}" \
--region "${GCP_REGION}" \
@@ -240,11 +243,12 @@ jobs:
# moved, the rollback target is the single thing an operator needs, and a summary that only
# appeared on success would be missing in exactly the run that needs it.
- name: Publish the rollout summary
if: ${{ always() && env.CANDIDATE_REVISION != '' && env.ROLLBACK_REVISION != '' }}
shell: bash
run: |
set -euo pipefail
{
echo '### Push gateway deployed'
echo '### Push gateway rollout'
echo
echo "Revision: \`${CANDIDATE_REVISION}\`"
echo
@@ -275,7 +279,7 @@ jobs:
# is not a failure to deploy, it is a live gateway that has to go back, so the traffic move
# is undone here rather than left to whoever reads the run.
- name: Roll traffic back to the previous revision
if: ${{ failure() && env.TRAFFIC_SHIFTED == 'true' }}
if: ${{ (failure() || cancelled()) && env.TRAFFIC_SHIFT_ATTEMPTED == 'true' }}
shell: bash
run: |
set -euo pipefail
@@ -290,6 +294,7 @@ jobs:
| jq -r '[.status.traffic[] | select((.percent // 0) > 0)]
| if length == 1 and .[0].percent == 100 then .[0].revisionName else empty end')"
test "${serving}" = "${ROLLBACK_REVISION}"
echo "TRAFFIC_ROLLED_BACK=true" >> "${GITHUB_ENV}"
{
echo
echo '### Push gateway rolled back'
@@ -302,8 +307,8 @@ jobs:
# SQL pool for nothing. Its tag comes off first, because Cloud Run refuses to delete a
# revision a traffic target still names, and clearing CANDIDATE_TAG makes the always() tag
# step below a no-op rather than a second failure.
- name: Delete the candidate revision that never took traffic
if: ${{ failure() && env.TRAFFIC_SHIFTED != 'true' }}
- name: Delete the rejected candidate revision
if: ${{ (failure() || cancelled()) && (env.TRAFFIC_SHIFT_ATTEMPTED != 'true' || env.TRAFFIC_ROLLED_BACK == 'true') }}
shell: bash
run: |
set -euo pipefail
.github/workflows/cloud-verify.yml
+1
View File
@@ -90,6 +90,7 @@ jobs:
--health-timeout 5s
--health-retries 10
env:
ORCA_PUSH_TEST_DATABASE_URL: postgres://relay_test:relay_test@127.0.0.1:5432/orca_relay_test
ORCA_RELAY_TEST_POSTGRES_URL: postgres://relay_test:relay_test@127.0.0.1:5432/orca_relay_test
steps:
- uses: actions/checkout@v4