Description
It's expected that sql INSERT is currently slower than kv puts because sql lacks partition pruning. But even on a single-partition table (this case should not be affected by lack of partition pruning) INSERT is 5x times slower than kv.
Let's investigate performance of sql INSERT.
Here is a result of local (on my laptop; 2,6 GHz 6-Core Intel Core i7 32GB RAM) benchmark:
Benchmark | (clusterSize) | (fsync) | (partitionCount) | Mode | Cnt | Score | Error | Units |
---|---|---|---|---|---|---|---|---|
InsertBenchmark.kvInsert | 1 | false | 1 | avgt | 20 | 0.153 | ± 0.019 | ms/op |
InsertBenchmark.sqlInsert | 1 | false | 1 | avgt | 20 | 0.657 | ± 0.056 | ms/op |
Attachments
Issue Links
- is related to
-
IGNITE-17699 Improve implementation of QueryTaskExecuter
- Open
-
IGNITE-20888 Sql. Use single row replication request when possible
- Open
-
IGNITE-21047 Sql. Avoid spamming execution tasks when possible
- In Progress
-
IGNITE-20887 Sql. Avoid using blocking api in sql threads
- Resolved
-
IGNITE-21018 Sql. Introduce cache for plan mapping
- Resolved
-
IGNITE-21019 Sql. Improve conversion to BinaryRow in UpdatableTableImpl
- Resolved