flare-foundation/flare-systems-deployment v1.7.0

on GitHub

Changed

• the indexer database is renamed from flare_ftso_indexer to fsp_indexer.
  Every service in the stack takes the name from docker-compose.yaml, so the
  stack needs no changes from you — but anything of your own that reads the
  indexer database directly must be repointed: backups, dashboards, reward
  calculation, ad-hoc queries.
• bumped c-chain-indexer image to v2.0.0 and switched it to mode = "fsp" with
  history_epochs = 0. The indexer now only collects what the FSP stack needs
  and resolves the contracts by name itself, so the config template no longer
  lists any collect_transactions / collect_logs entries and no longer sets
  db.history_drop, which fsp mode ignores. A fresh sync takes well under a
  minute instead of hours: it fully indexes only the last hour or so of blocks,
  and backfills the FSP events behind that. Retention is no longer a fixed
  42-day window either — history drop now deletes below two reward epochs before
  the current epoch's start, so the indexed range grows as the indexer runs and
  settles at roughly 7 to 10 days on Flare and Songbird, sliding forward with
  each epoch.
• the indexer database moved to a new indexer_data_v2 volume. The old
  indexer_data volume is left untouched as a rollback point.
• the services that read the indexer database — system-client, ftso-client,
  fdc-client and tee-relay-client — now wait for the indexer's /health to report
  200 before they start, via a compose healthcheck. Previously they came up
  against an empty database and worked through their own retry and backoff paths
  until data appeared. If you raise indexer.history_epochs, raise the
  healthcheck's start_period to match: the first backfill then takes
  proportionally longer, and compose will not start the dependent services once
  it marks the indexer unhealthy.
• populate_config.sh no longer appends a FlareTeeManager log filter to the
  indexer config. v2 collects the TeeInstructionsSent events tee-relay-client
  reads on every network where the contract is deployed, so the generated filter
  only duplicated a built-in one. FlareTeeManager is still resolved for the
  tee-relay-client config.

Upgrading

docker compose down
git fetch --tags
git checkout v1.7.0
./populate_config.sh
docker compose pull
docker compose up -d

• ./populate_config.sh is required: the indexer config template changed.
• the indexer starts from an empty database and resyncs in under a minute.
  /health reports 503 until it has, and the services that read the database wait
  for it, so expect them to start a minute or so after docker compose up -d.
• do not run docker compose down -v, docker volume prune -a or
  docker system prune --volumes during the upgrade: all three delete the old
  indexer_data volume you are keeping in order to roll back.
• once v2 runs fine, reclaim the space with
  docker volume rm <project>_indexer_data, where <project> is this
  directory's name. List the volumes first with docker volume ls.

Rolling back

docker compose down
git checkout v1.6.1
./populate_config.sh
docker compose pull
docker compose up -d

• the v1 indexer picks up the old indexer_data volume and resumes from where
  it stopped, then re-indexes the blocks produced while v2 was running. Allow
  catchup time in proportion to how long v2 ran.

Full Changelog: https://github.com/flare-foundation/flare-systems-deployment/blob/v1.7.0/CHANGELOG.md