Upgrade Notice
This release is intended for Mainnet (celestia) and is built from the same v10.x commit as v10.5.0-mocha and v10.5.0-corto. It is a minor, non-consensus-breaking update for nodes already running app version 10.
Upgrade existing nodes
Download the celestia-app archive for your platform and verify it against checksums.txt. Validators should also download the matching fibre archive. Source builds require Go 1.26.6; make install builds the multiplexer and make build-fibre-server builds Fibre.
Stop the node and, for validators, Fibre. Replace both binaries and restart with the existing homes and database. Confirm the node resumes syncing and signing, and that Fibre reports ready (see below).
Update configuration
Fibre gains a [health] table in server_config.toml. Add it and any other missing settings with the new binary:
celestia-appd config sync --home ~/.celestia-app --fibre-home ~/.celestia-fibre --dry-run
celestia-appd config sync --home ~/.celestia-app --fibre-home ~/.celestia-fibreUse your actual home directories. This preserves existing values and creates backups before changes. Missing files are skipped. The defaults (check_interval = "10s", probe_timeout = "3s", max_block_age = "2m") apply when the table is absent, so the sync is optional but recommended.
Validators: Fibre changes
-
Health and readiness endpoint. The Fibre server now exposes the standard gRPC health service on its client-facing port (default 7980). Use it for service monitoring and alerting; the log names the failing check and what to do when the server is not ready. Readiness does not gate uploads or downloads. See the health guide.
grpc_health_probe -addr=127.0.0.1:7980 -tls -tls-no-verify # readiness grpc_health_probe -addr=127.0.0.1:7980 -tls -tls-no-verify -service=fibre-liveness # liveness
-
Object storage requires HTTPS.
object_storage.endpointmust now be anhttps://URL; a plaintexthttp://endpoint fails validation at startup. Fibre also checks at startup that the configured bucket is reachable with the configured credentials and refuses to start otherwise. Review your object storage settings before restarting. Prune failures now also report corrupt shard markers. -
Idle connections close sooner. The server idle timeout drops from five minutes to 30 seconds and the client idle timeout is 20 seconds. Clients reconnect on the next request; custom clients must tolerate this.
-
An unexpected gRPC server exit now terminates the Fibre process so a supervisor can restart it, and shutdown publishes not-ready before draining.
Supported operating systems
See the canonical matrix in the release README.
- Tested in CI: Ubuntu 24.04 LTS (Noble Numbat) on x86_64.
- Prebuilt binaries provided: Linux (
amd64,arm64), macOS (amd64,arm64), including the separate Fibre server. arm64 binaries are cross-compiled and are not executed in CI. - Minimum glibc: 2.38 for Linux multiplexer builds. Ubuntu 22.04 and older are unsupported for these builds. The separate Fibre binary requires glibc 2.34.
- Changes since v10.4.0: none.
What's Changed
- feat(fibre): add WaitForAllUploads client option (backport #8034) by @mergify[bot] in #8073
- feat(fibre): [high] add gRPC health and readiness diagnostics to the fibre server (backport #8009) by @mergify[bot] in #8132
- test(docker-e2e): [medium] refresh mocha archive peers for genesis sync (backport #8134) by @mergify[bot] in #8136
- fix(fibre): [low] wait for the in-flight store health probe instead of failing fast (backport #8135) by @mergify[bot] in #8137
- fix(app): [high] align prepare proposal tx filtering with process proposal (backport #8193) by @mergify[bot] in #8223
- fix(fibre): [low] shorten idle connection timeouts (backport #8151) by @mergify[bot] in #8155
- fix(app): [high] branch ante state per tx in prepare proposal (backport #8192) by @mergify[bot] in #8225
- fix(multiplexer): [high] start on the app's committed app version (backport #8203) by @mergify[bot] in #8219
- fix(multiplexer): [medium] initialize nextAppVersion (backport #8190) by @mergify[bot] in #8221
- fix(fibre): [medium] backport object storage fixes to v10.x (#8012, #8086, #8090, #8089) by @rootulp in #8267
- fix(deps): [high] bump hyperlane-cosmos to v1.3.1 by @rootulp in #8291
Full Changelog: v10.4.0...v10.5.0