Skip to content

Establish whether the DataFusion 55 TPC-H off-heap bump is real, and changelog it if so #5704

Closed

Description

@andygrove

Describe the bug

The DataFusion 55 branch raises the off-heap size in CometTPCHQuerySuite from 2g to 4g:

conf.set(MEMORY_OFFHEAP_SIZE.key, "4g")

It is not established whether DataFusion 55 genuinely needs more memory for these queries, or whether this is a consequence of the peak_mem_used accounting change in apache/datafusion#22710 that also disabled two CometAggregateSuite metric tests.

Expected behavior

If DataFusion 55 really does need more off-heap for TPC-H, that is a user-visible change and belongs in the changelog: anyone who sized their off-heap against TPC-H would be under-provisioned after upgrading. If it is only an accounting artifact, the bump should be reverted along with the metric fix.

Additional context

Found while reviewing #5262. Related to the ignored aggregate metric tests.

Activity

  1. added
    enhancementNew feature or request
    priority:mediumFunctional bugs, performance regressions, broken features
    area:aggregationHash aggregates, aggregate expressions
    on Sep 5, 2026
  2. andygrove commented on Sep 5, 2026

    @andygrove
    MemberAuthor

    Tracked under #5709 ([EPIC] DataFusion 55 upgrade follow-ups).

  3. andygrove commented on Sep 15, 2026

    @andygrove
    MemberAuthor

    Resolved by #5904. The full SF1 CometTPCHQuerySuite passes on current main with the original 2 GiB off-heap budget, with both debug and release native libraries, so the increase to 4 GiB was not needed for this configuration and the suite is back at 2 GiB. With no user-visible change in memory requirements there is nothing to add to the changelog.

    One caveat from that PR: it did not measure the original DataFusion 55 upgrade commit or other workloads, so whether the bump was ever transiently necessary is unknown. That is historical at this point and not worth chasing separately. If a real memory regression on DataFusion 55 shows up elsewhere it should get its own issue with the workload attached.

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

    area:aggregationHash aggregates, aggregate expressionsenhancementNew feature or requestpriority:mediumFunctional bugs, performance regressions, broken features

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions