Interpret filter performance statistics
Filter performance
Use execution statistics in the NIM log to understand a filter's total time, its individual processing stages, and the number of affected records.
Each filter execution writes a statistics entry to the log. All duration values are measured in milliseconds (ms). Start with total, then compare the individual stages to find where the filter spends most of its time.
Read a statistics entryDirect link to Read a statistics entry
Filter stats fManagerEmployees (ms): total:5, query:4, resolve:0,
docs:1, sort:0, ordering:0, exclude:0, lookup:0, append:0 #21/0/1
| Field | Meaning |
|---|---|
total | Total time to execute the filter. |
query | Time to initialize parameters and variables and calculate the raw query result. |
resolve | Time to resolve result-document references from the raw result and dataset. |
docs | Time to compose the resulting documents. |
sort | Time to sort the returned documents. |
ordering | Time to determine included and excluded documents. |
exclude | Time to remove rows using an exclusion-column setting. |
lookup | Time to process configured lookups. |
append | Time to append results from other filters. |
The trailing #included/removed-by-lookup/excluded counts show how many documents NIM kept, removed through lookups, and excluded. In #21/0/1, NIM kept 21 documents, removed none through a lookup, and excluded one.
Investigate a slow filterDirect link to Investigate a slow filter
- Compare
totalacross multiple executions to determine whether the delay is consistent. - Identify the largest individual stage.
- Review the related configuration:
| If this stage is high… | Review… |
|---|---|
query or resolve | Relations, data volume, source collection, and the filter's starting table. |
sort or ordering | Sort criteria, duplicate handling, and returned row count. |
lookup | Lookup count, target data volume, and whether a relation can better suit the requirement. |
append | The appended filters and their result sizes. |
- Change one factor at a time, rerun the filter, and compare the new log entry.
Filter statistics help identify a likely bottleneck; they do not by themselves prove the root cause. Use the result preview and collection history alongside the log entry.