Skip to content
This repository was archived by the owner on Nov 21, 2026. It is now read-only.
This repository was archived by the owner on Nov 21, 2026. It is now read-only.

Native time-series arms: use the engine's default SHARDS, not a fixed 4 that flatters ArcadeDB #241

Closed

Description

@tae898

The native time-series arms create their type with a fixed SHARDS 4. Upstream's answer on ArcadeData#9166 and the time-series docs say to leave SHARDS at its default (the async worker count, 11 under the benchmark cpuset), and a round 2 best-practice probe found the fixed 4 favours ArcadeDB: the 12 h hourly aggregate p50 was 0.505 ms at 4 shards and 1.007 ms at the default 11 (ratio of medians 1.99, bootstrap 95 percent interval 1.54 to 2.63, exact Mann-Whitney p 0.008, five runs per arm, JDK 25, official 26.10.1 jars, identical answers and PROFILE counters, ingest and newest-reading latency unchanged). Round 1 had compared only 3 against 4 shards and saw no effect.

Take the setting that does not flatter ArcadeDB: omit the SHARDS clause, record ts_shards as 0 and the engine-reported ts_shard_count_effective, keep TS_SHARDS=4 as the way to reproduce the old arm. This lands before the time-series stage starts on mini, so no rows are split. The probe is Java on laptop E cores and JDK 25 only, so the mini rows are the measurement; JDK 21 was not run.

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

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions