Repository navigation
v2.5.5
HexDB 2.5.5 stops replicas from reporting a reachable Overseer as unreachable, lets sliced replicas pass on live Server-Sent Events feeds, and gives the admin UI more room.
Fixed
- "This hex is a replica and couldn't reach the Overseer" with nothing wrong. A hex closes a keep-alive connection that has been idle for
limits.header_timeout_seconds(10 by default), but the client hexes use to talk to each other kept idle connections for 90 seconds. Now and then a replica reused a connection just as the Overseer closed it, and a forwarded request failed with a 502. Sliced replicas forward reads too, so an admin UI page polling every few seconds hit it often. Connections between hexes are now dropped after half the header timeout, and a forwarded read (GETorHEAD) is sent again if its connection breaks mid-request, as writes with an idempotency key already were. - Live stream views on sliced replicas. A sliced replica refused every live feed with 421, naming the Overseer's internal address, so the admin UI's live view of a stream never updated through one. Server-Sent Events feeds (
/changes/stream,/streams/{name}/subscribe) now stream through from the Overseer like any other forwarded answer. WebSocket feeds (/changes/ws,/streams/{name}/ws) still answer 421sliced_replica.
Admin UI
- The introductions at the top of Streams, Queues, Functions, Roles, Audit, Settings and the other pages use the full width instead of stopping halfway.
- The Schemas page has the same padding as every other page, a wider tessellation list and more space around the editor.
Upgrading
Restart the hexes one at a time, as usual: replicas first, the Overseer last.
Full Changelog: v2.5.4...v2.5.5