During our regular performance measurement we realized a performance degradation when sending CRUD operations through NCMP.

Current behavior

CPS hash: 3.3.8-cc8b8257dcceaa03bf46a6615b6347a5573906a3

Querying 5k data through 24 threads on ncmp/v1/ch/${cmHandleHash}/data/ds/ncmp-datastore:passthrough-operational interface:

1.  4.47 s response time per operation. Configuration was 2 instances with SPRING_DATASOURCE_HIKARI_MAXIMUMPOOLSIZE set to 40.
2. 4.46 s response time per operation. Configuration was 2 instances with SPRING_DATASOURCE_HIKARI_MAXIMUMPOOLSIZE set to 80.
3. 10.05 s response time per operation. Configuration was 1 instance with SPRING_DATASOURCE_HIKARI_MAXIMUMPOOLSIZE set to 80.

Previous behavior

CPS hash: 3.3.7-c534c78e2d960c11875b5cc88ec605a78ecb1624

Querying 5k data through 24 threads on ncmp/v1/ch/${cmHandleHash}/data/ds/ncmp-datastore:passthrough-operational interface:

1.  2.37 s response time per operation. Configuration was 2 instances with SPRING_DATASOURCE_HIKARI_MAXIMUMPOOLSIZE set to 40.
2. 2.39 s response time per operation. Configuration was 2 instances with SPRING_DATASOURCE_HIKARI_MAXIMUMPOOLSIZE set to 80.
3. 2.43 s response time per operation. Configuration was 1 instance with SPRING_DATASOURCE_HIKARI_MAXIMUMPOOLSIZE set to 80.

Reproduction

• Run a cmHandle discovery with 20k nodes, get them to READY state
• Run data queries on 24 threads against the ncmp/v1/ch/${cmHandleHash}/data/ds/ncmp-datastore:passthrough-operational interface.
• Check for response times during the performance measurement

Test environment:

• Real cluster
• No mocks in the system
• Postgres database
• Real model adapter, same version for every test