Promote deployed subgraphs #375
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Promote deployed subgraphs | |
| # A subgraph can take hours to index, far longer than a job may run, so the deployment | |
| # workflows do not wait for it: they flag the version they deployed in a repository | |
| # variable named SUBGRAPH_PENDING_<ENV>_<NETWORK>, and this schedule picks it up. | |
| # | |
| # Roughly once an hour it checks each flagged deployment and, as soon as it is fully | |
| # indexed, moves the 'latest' tag onto it and deletes the version the tag was pointing | |
| # to. The flag is then cleared - as it is when the indexing failed, which fails the job | |
| # so the failure is reported. Nothing flagged means nothing to do, and the run costs a | |
| # few seconds. | |
| # | |
| # A deployment that neither indexes nor fails is watched for PROMOTION_DEADLINE_HOURS and | |
| # given up on after that: a version nobody is ever going to promote gets reported once, | |
| # rather than checked over and over forever. | |
| # | |
| # To watch a deployment again - typically after a failed indexing has been fixed and | |
| # restarted - set the variable back to the deployed version (Settings > Secrets and | |
| # variables > Actions > Variables) and run this workflow. | |
| on: | |
| schedule: | |
| # Once an hour, at :25 rather than on the hour, and best-effort rather than on time: | |
| # GitHub dispatches scheduled runs when it gets to them - tens of minutes late here, | |
| # and later still on the hour, where every other repository's cron lands - and a tick | |
| # still undelivered when the next one is due is dropped rather than queued. Asking for | |
| # more than one an hour buys nothing: the ':20 and :50' this replaced was delivered | |
| # about once every three hours, one tick in six. Nothing here needs a tick at a set | |
| # time, and a missed one only leaves the flag to the following run. | |
| # | |
| # Scheduled workflows only ever run from the default branch, and GitHub disables them | |
| # after 60 days without any repository activity, which is what stopped the | |
| # 'pin-to-pinata' schedule on 2026-08-04. | |
| - cron: "25 * * * *" | |
| workflow_dispatch: | |
| permissions: | |
| contents: read | |
| env: | |
| # Indexing a new version takes hours, and indexing a busy chain from scratch can take | |
| # most of a day, so this is deliberately wide: it is the point where a deployment is | |
| # declared never going to make it, not an indexing budget. | |
| PROMOTION_DEADLINE_HOURS: "24" | |
| jobs: | |
| list-pending: | |
| name: List the deployments waiting for promotion | |
| runs-on: ubuntu-latest | |
| timeout-minutes: 5 | |
| outputs: | |
| pending: ${{ steps.list.outputs.pending }} | |
| steps: | |
| # No checkout and no install: on the vast majority of runs nothing is flagged and | |
| # this is all the schedule does | |
| - id: list | |
| name: Read the pending deployment flags | |
| env: | |
| SUBGRAPH_PENDING_TESTING_SEPOLIA: ${{ vars.SUBGRAPH_PENDING_TESTING_SEPOLIA }} | |
| SUBGRAPH_PENDING_TESTING_BASE: ${{ vars.SUBGRAPH_PENDING_TESTING_BASE }} | |
| SUBGRAPH_PENDING_STAGING_SEPOLIA: ${{ vars.SUBGRAPH_PENDING_STAGING_SEPOLIA }} | |
| SUBGRAPH_PENDING_STAGING_BASE: ${{ vars.SUBGRAPH_PENDING_STAGING_BASE }} | |
| SUBGRAPH_PENDING_PRODUCTION_ETHEREUM: ${{ vars.SUBGRAPH_PENDING_PRODUCTION_ETHEREUM }} | |
| SUBGRAPH_PENDING_PRODUCTION_BASE: ${{ vars.SUBGRAPH_PENDING_PRODUCTION_BASE }} | |
| run: | | |
| set -euo pipefail | |
| # "<boson env>|<network>|<label>", one per Goldsky hosted subgraph. Doubles as | |
| # the allow list a promotion job is built out of: the env key and the variable | |
| # name are derived from the first two fields, never spelled out a second time. | |
| subgraphs=( | |
| "testing|sepolia|Sepolia (testing)" | |
| "testing|base|Base Sepolia (testing)" | |
| "staging|sepolia|Sepolia (staging)" | |
| "staging|base|Base Sepolia (staging)" | |
| "production|ethereum|Ethereum (production)" | |
| "production|base|Base (production)" | |
| ) | |
| entries="" | |
| malformed="" | |
| for subgraph in "${subgraphs[@]}"; do | |
| IFS="|" read -r boson_env network label <<< "$subgraph" | |
| key="${boson_env^^}" | |
| variable="SUBGRAPH_PENDING_${key}_${network^^}" | |
| version="${!variable:-}" | |
| if [ -z "$version" ]; then | |
| continue | |
| fi | |
| # The version goes into a JSON matrix and onto a command line, so only ever | |
| # trust one that still looks like the version the deployment flagged. Skip the | |
| # subgraph rather than stop here: the other networks have a deployment of | |
| # their own waiting, and this run is failed at the end anyway. | |
| if [[ ! "$version" =~ ^[A-Za-z0-9.+-]+$ ]]; then | |
| echo "::error::Repository variable $variable does not hold a version: $version" | |
| malformed="$malformed $variable" | |
| continue | |
| fi | |
| entry='{"boson_env":"'"$boson_env"'","network":"'"$network"'","key":"'"$key"'","label":"'"$label"'","variable":"'"$variable"'","version":"'"$version"'"}' | |
| entries="$entries${entries:+,}$entry" | |
| done | |
| echo "pending=[$entries]" | tee -a "$GITHUB_OUTPUT" | |
| if [ -n "$malformed" ]; then | |
| echo "Not holding a version:$malformed" | |
| exit 1 | |
| fi | |
| promote: | |
| name: Promote ${{ matrix.label }} subgraph | |
| needs: list-pending | |
| # A malformed variable fails 'list-pending' only once it has listed the sound ones, | |
| # which are promoted all the same | |
| if: ${{ !cancelled() && needs.list-pending.outputs.pending != '' && needs.list-pending.outputs.pending != '[]' }} | |
| runs-on: ubuntu-latest | |
| timeout-minutes: 20 | |
| strategy: | |
| # A subgraph that failed to index must not stop the others from being promoted | |
| fail-fast: false | |
| matrix: | |
| include: ${{ fromJSON(needs.list-pending.outputs.pending) }} | |
| # Two runs must never move the same 'latest' tag at the same time. A tick that is | |
| # still running when the next one fires is the one holding the fresher answer, so let | |
| # it finish rather than cancel it. | |
| concurrency: | |
| group: subgraph-promote-${{ matrix.boson_env }}-${{ matrix.network }} | |
| cancel-in-progress: false | |
| steps: | |
| - uses: actions/checkout@v3 | |
| with: | |
| token: ${{ secrets.BSNORG_ACTIONS_SECRET }} | |
| - uses: ./.github/actions/setup-subgraph | |
| with: | |
| # Tagging an already deployed version runs a single script out of the subgraph | |
| # workspace: it needs neither the other workspaces nor the built subgraph | |
| build: "false" | |
| workspace: packages/subgraph | |
| # Reports 'promoted', 'pending' or 'failed' on its 'outcome' output. Anything else | |
| # going wrong - Goldsky being unreachable, say - leaves the output unset, so the | |
| # deployment stays flagged and is checked again on the next run. | |
| - id: promote | |
| name: Promote subgraph | |
| run: cd ./packages/subgraph && npm run promote:${{ matrix.boson_env }}:${{ matrix.network }} -- --deployed-version=${{ matrix.version }} | |
| env: | |
| GOLDSKY_API_KEY: ${{ secrets[format('GOLDSKY_API_KEY_{0}', matrix.key)] }} | |
| - name: Clear the pending deployment flag | |
| # Either the 'latest' tag has been moved, or the version will never index: both | |
| # end the watch, and a failed indexing has already failed the step above | |
| if: ${{ !cancelled() && contains(fromJSON('["promoted", "failed"]'), steps.promote.outputs.outcome) }} | |
| env: | |
| GH_TOKEN: ${{ secrets.BSNORG_ACTIONS_SECRET }} | |
| VARIABLE: ${{ matrix.variable }} | |
| VERSION: ${{ matrix.version }} | |
| run: | | |
| set -euo pipefail | |
| # A deployment that landed while this job was running has flagged the version it | |
| # deployed: clearing the flag now would drop that one silently | |
| flagged=$(gh api "repos/$GITHUB_REPOSITORY/actions/variables/$VARIABLE" --jq ".value" 2>/dev/null || true) | |
| if [ "$flagged" != "$VERSION" ]; then | |
| echo "::notice::$VARIABLE holds '$flagged' and no longer '$VERSION' - leaving it to the next run" | |
| exit 0 | |
| fi | |
| gh variable delete "$VARIABLE" --repo "$GITHUB_REPOSITORY" | |
| - name: Give up on a deployment that never promotes | |
| # Whatever left the flag in place: a version still indexing, or a step that failed | |
| # before reporting an outcome at all - a missing API key, a version deleted by | |
| # hand on Goldsky. Without this, that flag is picked up again on every run, | |
| # forever, and nothing ever says so. | |
| if: ${{ !cancelled() && !contains(fromJSON('["promoted", "failed"]'), steps.promote.outputs.outcome) }} | |
| env: | |
| GH_TOKEN: ${{ secrets.BSNORG_ACTIONS_SECRET }} | |
| VARIABLE: ${{ matrix.variable }} | |
| VERSION: ${{ matrix.version }} | |
| LABEL: ${{ matrix.label }} | |
| run: | | |
| set -euo pipefail | |
| flag=$(gh api "repos/$GITHUB_REPOSITORY/actions/variables/$VARIABLE" 2>/dev/null || true) | |
| if [ -z "$flag" ]; then | |
| echo "::notice::$VARIABLE is gone - nothing left to watch" | |
| exit 0 | |
| fi | |
| flagged=$(jq -r ".value" <<< "$flag") | |
| if [ "$flagged" != "$VERSION" ]; then | |
| echo "::notice::$VARIABLE holds '$flagged' and no longer '$VERSION' - leaving it to the next run" | |
| exit 0 | |
| fi | |
| # The flag is written once, by the deployment that raised it, so how long it has | |
| # been waiting is how long ago it was last written | |
| flagged_at=$(jq -r ".updated_at" <<< "$flag") | |
| waited_hours=$(( ( $(date -u +%s) - $(date -u -d "$flagged_at" +%s) ) / 3600 )) | |
| if [ "$waited_hours" -lt "$PROMOTION_DEADLINE_HOURS" ]; then | |
| echo "::notice::$LABEL $VERSION has been waiting for promotion for ${waited_hours}h - checked again on the next run" | |
| exit 0 | |
| fi | |
| echo "::error::$LABEL $VERSION has been waiting for promotion for ${waited_hours}h, past the ${PROMOTION_DEADLINE_HOURS}h deadline - giving up on it. Once the deployment is fixed, set $VARIABLE back to a deployed version to watch it again." | |
| gh variable delete "$VARIABLE" --repo "$GITHUB_REPOSITORY" | |
| exit 1 |