fix!: fix regression caused by unbalanced partitions and splitting ranges (#5090)

* feat: assign partition ranges by rows

* feat: balance partition rows

* feat: get uppoer bound for part nums

* feat: only split in non-compaction seq scan

* fix: parallel scan on multiple sources

* fix: can split check

* feat: scanner prepare by request

* feat: remove scan_parallelism

* docs: upate docs

* chore: update comment

* style: fix clippy

* feat: skip merge and dedup if there is only one source

* chore: Revert "feat: skip merge and dedup if there is only one source"

Since memtable won't do dedup jobs

This reverts commit 2fc7a54b11.

* test: avoid compaction in sqlness window sort test

* chore: do not create semaphore if num partitions is enough

* chore: more assertions

* chore: fix typo

* fix: compaction flag not set

* chore: address review comments
This commit is contained in:
Yingwen
2024-12-09 20:50:57 +08:00
committed by GitHub
parent 1b642ea6a9
commit 2fcb95f50a
19 changed files with 560 additions and 309 deletions

View File

@@ -492,12 +492,6 @@ experimental_write_cache_ttl = "8h"
## Buffer size for SST writing.
sst_write_buffer_size = "8MB"
## Parallelism to scan a region (default: 1/4 of cpu cores).
## - `0`: using the default value (1/4 of cpu cores).
## - `1`: scan in current thread.
## - `n`: scan in parallelism n.
scan_parallelism = 0
## Capacity of the channel to send data from parallel scan tasks to the main task.
parallel_scan_channel_size = 32