Skip to content

OOM after large amount of update and restart #1910

@JaySon-Huang

Description

@JaySon-Huang

Steps to reproduce:

  1. Load a TPCH-50 lineitem and its TiFlash replica with clustered index ("clustered index" maybe not necessary)
  2. Stop TiFlash node
  3. Update 1% or 10% by executing alter table lineitem add column l_num bigint, update lineitem set l_num = l_partkey / 100 where l_partkey % 100 = 1
  4. Start TiFlash node

After TiFlash starts, the memory usage grows to 120GB+ and is killed by the OS system.
There are no rows written to storage, we assume that no committed data can be scanned from the Raft log applied.
The default CF size of proxy shown in Grafana is about 9.5GB.

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions