trafficserver icon indicating copy to clipboard operation
trafficserver copied to clipboard

Adds predicting server failures experimental plugin

Open IsaacHughes74 opened this issue 3 years ago • 5 comments

Monitors data rate and server connections to predict origin server failure. Utilizes "death cross" of data rate, where data rate EMA with window size of 5s falls below data rate EMA with window size of 25s, while server connections increase. Slope and EMA calculations are maintained by a background thread periodically calculating statistics. The number of server connections is calculated by incrementing at TS_EVENT_HTTP_SEND_REQUEST_HDR and decrementing at TS_EVENT_HTTP_TXN_CLOSE, and the data rate is calculated by dividing the content length of the response by the time between TS_MILESTONE_SERVER_BEGIN_WRITE and TS_MILESTONE_SERVER_CLOSE.

Moving forward:

  • Add configurability - Remove hardcoded EMA period calculations of 5s and 25s
  • Expand - Include additional metrics to expand scope

IsaacHughes74 avatar Aug 04 '22 23:08 IsaacHughes74

[approve ci]

ezelkow1 avatar Aug 05 '22 00:08 ezelkow1

Would be nice to add documentation

shukitchan avatar Aug 12 '22 00:08 shukitchan

[approve ci autest]

shukitchan avatar Aug 12 '22 00:08 shukitchan

[approve ci autest]

bryancall avatar Aug 29 '22 23:08 bryancall

[approve ci]

bryancall avatar Aug 29 '22 23:08 bryancall

This pull request has been automatically marked as stale because it has not had recent activity. Marking it stale to flag it for further consideration by the community.

github-actions[bot] avatar Feb 13 '23 02:02 github-actions[bot]