You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 8ae0262
Browse filesBrowse the repository at this point in the historyBrowse files
chore(relay): deploy with the Alchemy CLI and publish client config through an Action
infra/relay/scripts/deploy.ts reimplemented the CLI's bootstrap against
Alchemy internals (LoggingCli, ArtifactStore, AlchemyContext, Plan, Apply)
to add four things: a confirm prompt, writing the relay URL and tracing
config to the repo-root .env, GitHub Actions outputs, and a read-state mode
for the release job. The bootstrap broke on every Alchemy bump.
- Deploy is `alchemy deploy`; every flag the script exposed exists on the
CLI (--dry-run, --force, --stage, --yes, --adopt, --env-file, --no-input).
- The .env write is a PublishClientConfig Action in the stack. Its input is
the stack's client outputs, so it runs when one of them changes and is
skipped on a no-op deploy. Alchemy hashes JSON.stringify of the input and
Redacted stringifies opaquely, so a digest of the two tokens rides along
to make rotations re-run it. T3CODE_RELAY_CLIENT_CONFIG_ENV redirects the
write; CI points it at a temp file.
- The release job reads prod's outputs with `alchemy state read
T3CodeRelay/prod/output` and picks the tracing fields with jq (redacted
values persist as {"__redacted__": value}).
- The deploy-relay commit status loses its applied-vs-no-op wording; the
CLI does not report that and the job result is the signal that matters.
Co-Authored-By: Claude Fable 5 <[email protected]>
0 commit comments