| Title | eventcnv's bucket aggregation is broken |
| Status | closed |
| Priority | optional |
| Assigned user | Gareth Rees |
| Organization | Ravenbrook |
| Description | If you request that statistics are aggregated into buckets by passing the -b option to eventcnv, the results are bogus. |
| Analysis | This code relies on events being processed in order of their eventTime: /* Output bucket, if necessary, and update counters */ if (bucketSize != 0 && eventTime >= bucketLimit) { reportBucketResults(bucketLimit-1); clearBucket(); do { bucketLimit += bucketSize; } while (eventTime >= bucketLimit); } but events may appear out of order in the telemetry file, so this algorithm is bogus. The events need to be sorted before aggregation into buckets (or, alternatively, all events need to be processed before any bucket is output). For the moment I have dealt with this by not documenting the -b option to eventcnv. |
| How found | inspection |
| Evidence | See analysis. |
| Observed in | 1.110.0 |
| Created by | Gareth Rees |
| Created on | 2012-10-21 16:35:36 |
| Last modified by | Gareth Rees |
| Last modified on | 2012-10-23 14:28:22 |
| History | 2012-10-21 GDR Created. |
| Change | Effect | Date | User | Description |
|---|---|---|---|---|
| 180028 | closed | 2012-10-23 14:28:22 | Gareth Rees | Remove reporting features from eventcnv, as discussed with RB. Eventcnv was always intended to be a simple tool that just translates events, not a report generator. In particular: * Remove -e option and associated event specification parsing. * Remove -b option and bucket statistics. * Don't intern labels (always print addresses). * Remove -v option (events are always output). * If -h or -? is specified, exit after printing help message. * Help text refer to the "Telemetry" section of reference manual. |