hyperledger / fabric-x-committer

89%

main: 90%

LAST BUILD ON BRANCH split-tablets

branch: split-tablets

CHANGE BRANCH

x

Reset

• split-tablets
• apt-ci-fix
• bft-delivery
• ci-gotestsum
• commit-config-immediately
• config-docs-yaml
• connection-test-dep
• coveralls
• cryptogen-recreate-support
• default-test-node-startup-setup
• deliver-check-block-order
• deliver-seek-tls-cert
• doc-loadgen-artifacts
• dynamic-root-ca
• enable-pprof
• expose-ledger-api
• fabric-logging
• fix-connection-reset
• fix-container-image
• fix-container-test-macos
• fix-db-test
• fix-falky-svmupdate-test
• fix-flaky-conn-metrics
• fix-flaky-loadgen-coord
• fix-master-tserver-rf
• fix-port-container-test
• fix-release-path
• fix-sign
• fix-test-node-with-insecure
• fix-waiting-tx-count
• go-fix
• ledger-service-wait-behaviour
• lifecycle-policy
• limit-block-apis
• limit-inputs-block-apis
• loadgen-remove-ctx
• main
• make-fixes
• metrics-tls
• migrate-to-postgres-official-image
• minimize-logs
• mock-orderer-features
• mock-orderer-stream-holder
• namespace-policy-docs
• optimize-map-block
• pgx5
• postgres-renaming
• qs-pen-test-fix
• rate-limit-sidecar
• root-ca-workaround-fix
• sidecar-block-store-code-fix
• sidecar-limit-active-txids-notifier
• sidecar-remove-chan-id
• sustain-config
• sync-block
• template-tls-per-endpoint
• test-coverage-improve
• test-coverage-tests
• test-limiter-max-wait
• test-minor-fixes
• test-query-status
• test-tls-flatten
• testcrypto-refactor
• tls-docs-update
• update-to-latest-common
• update-to-yugabytedb-latest-lts
• upgrade-pg
• worktree-notification-docs

Build # 21869758251

Build Type

Pull #311

github

Committed by cendhu

Commit Message

[vc] split tables into predefined number of tablets

When a new table is created, YugabyteDB starts with a single
tablet that handles ALL initial writes. Auto-splitting only kicks in after
the tablet reaches a size threshold (~1 GB by default), meaning the first
N transactions all funnel through one tablet on one tserver — creating a
severe bottleneck during the critical warm-up phase.

By pre-splitting tables at CREATE TABLE time (e.g., SPLIT INTO 3 TABLETS
for a 3-node cluster), data is hash-distributed across all tservers from
the very first write. This converts cold-start throughput from
single-tserver-bound to cluster-wide from block 1.

This applies to all three table types: metadata, tx_status, and
per-namespace state tables (ns_*). The setting auto-detects the DB backend
and is silently ignored on PostgreSQL, which has no tablet concept.

Signed-off-by: Senthilnathan <cendhu@gmail.com>

Pull Request Pull Request #311: [vc] split tables into predefined number of tablets

Run Details

31 of 37 new or added lines in 2 files covered. (83.78%)

4 existing lines in 3 files now uncovered.

9089 of 10238 relevant lines covered (88.78%)

1.29 hits per line