Slow Neo4j queries 4,808
Individual slow Cypher on test26 over the last 3d —
by elapsed time, with the CPU / wait / page-hit profile so you can tell genuine compute from
lock-wait. Click a column to sort. Endpoint latency is a whole-request concern — see Calls.
What the slowest ones have in common
The slowest 300 statements in this window (over 1.32 s) compared against the other 2700. Ranked by how much of the slow set each attribute accounts for — a value with huge lift covering 2% is a curiosity, one at 3× covering most of them is the lead.
| When | Service | Elapsed ▾ | CPU | Wait | Page hits | R/W | Query |
|---|---|---|---|---|---|---|---|
| 2d ago | docsvc | 100 ms | 100 ms | 0 ms | 57k | read | None |
| 2d ago | execsvc | 100 ms | 100 ms | 0 ms | 641k | read | None |
| 2d ago | docsvc | 100 ms | 15 ms | 0 ms | 16 | write | None |
| 2d ago | docsvc | 100 ms | 19 ms | 0 ms | 0 | write | None |
| 2d ago | virtual-assistant | 100 ms | 5 ms | 0 ms | 0 | read | None |
| 2d ago | docsvc | 100 ms | 100 ms | 0 ms | 142 | read | None |
| 2d ago | virtual-assistant | 100 ms | 100 ms | 0 ms | 107k | write | None |
| 2d ago | docproc | 100 ms | 100 ms | 0 ms | 148 | read | None |
4,808 slow queries · 100/page · Elapsed total; CPU ≈ elapsed (red) = genuine compute — check Page hits (millions = full scan / missing index); high Wait (amber) = blocked on locks. A — profile is an older row awaiting backfill (new rows are profiled at ingest).