Metrics (P90/P100)
Hud measures how long every endpoint, queue, custom flow and function takes. It counts every call and every error, and samples durations. You'll see durations as an average, P90 or P100.
Average, P90 and P100
| Metric | What it is | What it's good for |
|---|---|---|
| Average | Total time divided by the number of calls | A rough sense of speed. One very slow call can pull it up a lot. |
| P90 | 90% of calls take this long or less | How fast it is for almost everyone, without the rare outliers |
| P100 | The slowest call in the time frame | Worst cases, such as timeouts |
Example
Ten requests take these times, in ms:
| Request | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
|---|---|---|---|---|---|---|---|---|---|---|
| Time (ms) | 100 | 110 | 120 | 130 | 140 | 150 | 160 | 170 | 180 | 2,000 |
- Average: 426 ms. The one slow request pulls it far above what most requests took.
- P90: 180 ms. Nine of the ten requests took this long or less, so it shows typical speed.
- P100: 2,000 ms. The slow request itself.
Where Hud uses each
- Slowdown rules: 1 Week Degradation, 1 Day Degradation, Latency Spike and Post-Deploy Degradation compare P90 duration with the flow's normal. By default they skip flows whose P90 is under 150 ms. See Slowdowns over time and Post-Deploy Issues.
- SLO rules: SLO Breach alerts compare a percentile you choose (P90, P95, P99 or P100) with your SLO. The default is P100, so a single slow call counts. See SLO rules.
Questions about a metric? Chat with our support team or email us at [email protected]
Updated 3 days ago
What’s Next
Did this page help you?