Skip to content

A server page's first visit aborts and re-sends its in-flight panel reads when the shell's fleet read returns #9220

Closed

Description

@erikdarlingdata

Claude posting for Erik Darling.

What

Build 3.13.0-nightly.20261121.714, web dashboard. #8730 (closed) described this on a PostgreSQL target's Overview: the page starts its panel reads before the fleet roll-up answers (#8554), and when the fleet answer arrives the page restarts those panels, so the browser aborts the reads in flight and sends them again. #8733 closed it by making the ui-smoke check open server pages as a second visit. The page itself was not changed, and the race is still there, now on a SQL Server target too.

With a fresh browser context for each first visit, every store reachable and no read blocked, then one reload of each page:

  • Queries tab of a busy SQL Server target, first visit: four open-phase reads ended net::ERR_ABORTED and were sent again: the 24-hour query duration trend, the 24-hour procedure duration trend, top queries by CPU and top procedures by CPU. All four aborted within a hundredth of a second of the last of the shell's /api/fleet reads returning (the shell sends that read three times on a first visit).
  • Overview of the same server: more than half of the first visits had one or two aborted reads (memory trend, file I/O trend, CPU utilization), each ending just after the last /api/fleet read.
  • The reload of each page: no aborted reads.
  • On the previous nightly, SQL Server first visits showed none. A slower first visit (a slower store) makes the window wider: the earlier build of this nightly, under a heavy store write load, aborted three to four reads per visit.

An aborted read costs the user nothing visible on its own, but the page sends it twice, and on the Queries tab the reads that abort are the heavy ones. Whether the store finishes an aborted read anyway was not measured.

Proposed

  • Keep the reads already in flight when the fleet answer arrives. A panel that does not depend on the fleet answer is not rebuilt; the parts that do depend on it update in place.
  • Restore the strict first-visit check in ui-smoke (open a server's Overview and Queries tab as a first visit and fail on any aborted read), so ui-smoke: PostgreSQL server pages open as a second visit (#8730) #8733's workaround stops hiding the race.

Tests

  • ui-smoke: a first visit to a server's Overview and Queries tab, for a SQL Server and a PostgreSQL target, with a fleet answer slower than the first panel reads: zero ERR_ABORTED reads, and each panel's read is sent once.
  • A fleet answer that returns before any panel read starts, and one that returns after every panel read finished, give the same result.
  • The page's figures after the fleet answer match what they were before the change.

Activity

  1. added
    in-progressActively being worked by a local session or its agents (PR open or in flight)
    on Nov 21, 2026
  2. erikdarlingdata commented on Nov 21, 2026

    @erikdarlingdata
    OwnerAuthor

    Claude posting for Erik Darling. Fixed by #9226, merged to dev. It ships in the next release.

  3. removed
    in-progressActively being worked by a local session or its agents (PR open or in flight)
    on Nov 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions