Changelog

7.10.8.41 — September 2026

Services

  • Migrated the source-code and build platform onto faster hardware with solid-state storage, cutting write latency threefold, with a three and a half minute interruption.
  • Added longer transfer timeouts to the source-code platform's entry point so large repository operations no longer break off midway.

Stability

  • Fixed services across the platform being unable to start for hours after an overnight maintenance restart, caused by stale storage left behind on two thirds of the machines.
  • Fixed storage redundancy being unable to rebuild itself, where too many simultaneous rebuilds collided and repeatedly undid each other's work.
  • Fixed four services left without their data, recovering two in place, rebuilding one from its live replica and restoring the rest from that morning's backup.
  • Upgraded the distributed storage platform on both clusters to its current release.
  • Fixed every single-sign-on protected application looping endlessly at login, caused by an identity release that discards forwarded request headers arriving from untrusted sources.
  • Fixed the authentication gateways staying on the prior release after their server was upgraded, leaving the corrective fix undeployed until each gateway was forced to rebuild.
  • Fixed three empty namespaces created by a single mistyped character in a deployment target.
  • Fixed a build runner authenticating with the shared default account instead of a dedicated identity of its own.
  • Fixed one connection in five to the platform hanging for five minutes instead of loading, caused by a host task limit that killed web workers and left their listening slots unanswered.
  • Upgraded a build runner to match the source-code platform's version.
  • In progress bringing one hundred repositories onto the current pipeline template, forty complete.

Documentation

  • Documentation of the platform hardware migration, covering the procedure, the verification tests and the lessons learned.
  • Documentation of the connection-hang incident, covering the diagnosis, the host limit behind it and the checks that reported healthy throughout.

Tracks in this release: Services · Stability · Documentation

Shipped Features

1
Services would not start for hours after an overnight maintenance restartComplete
1
Storage redundancy could not rebuild itself after the outageComplete
1
Four services lost access to their data during the storage faultComplete
1
Every single-sign-on protected application redirects in a loopComplete
1
Sign-on gateways stayed on the old release after the server was upgradedComplete
1
Three empty namespaces created by a mistyped deployment targetComplete
1
A build runner authenticated with the shared default accountComplete
1
One connection in five hung for five minutes instead of loadingComplete
1
Move the source-code and build platform onto solid-state storageComplete
1
Large repository transfers break off midwayComplete
1
Document the platform hardware migrationComplete
1
Bring all repositories onto the current pipeline templateIn Progress
1
Document the connection-hang incident and its causeComplete
1
Upgrade the distributed storage platform on both clusters to its current releaseComplete
1
Upgrade a build runner to match the source-code platform's versionComplete