[SERVER-47644] Enable Latency Stats in FTDC Created: 17/Apr/20  Updated: 29/Oct/23  Resolved: 07/May/20

Status: Closed
Project: Core Server
Component/s: Diagnostics
Affects Version/s: 3.6.17, 4.2.6, 4.0.18, 4.4.0-rc1
Fix Version/s: 4.4.0-rc5, 4.7.0

Type: Improvement Priority: Major - P3
Reporter: Kevin Arhelger Assignee: Mark Benvenuto
Resolution: Fixed Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Backports
Related
Backwards Compatibility: Fully Compatible
Backport Requested:
v4.4
Sprint: Security 2020-05-04, Security 2020-05-18
Participants:

 Description   

SERVER-25180 suggests an option such as diagnosticDataCollectionEnableLatencyHistograms to enable FTDC collection of latency histograms. However this does not seem to be implemented.

Adding histogram buckets <= slowms would allow for a better understanding of latency spikes and response time distributions.



 Comments   
Comment by Githook User [ 08/May/20 ]

Author:

{'name': 'Mark Benvenuto', 'email': 'mark.benvenuto@mongodb.com', 'username': 'markbenvenuto'}

Message: SERVER-47644 Enable Histogram Latency Stats in FTDC

(cherry picked from commit 049b10ad31be3233130dbf4313b6be4a3ecd47c9)
Branch: v4.4
https://github.com/mongodb/mongo/commit/07b6549afa70178570b0af93f536db1b8c5643a3

Comment by Githook User [ 07/May/20 ]

Author:

{'name': 'Mark Benvenuto', 'email': 'mark.benvenuto@mongodb.com', 'username': 'markbenvenuto'}

Message: SERVER-47644 Enable Histogram Latency Stats in FTDC
Branch: master
https://github.com/mongodb/mongo/commit/049b10ad31be3233130dbf4313b6be4a3ecd47c9

Comment by Bruce Lucas (Inactive) [ 04/May/20 ]

Not a strong opinion. Option 3 would mean that the last bucket would have a different meaning from the bucket of the same name in the unfiltered histogram. On the other hand it conveys useful information, and I think it is currently the case that the last bucket in the unfiltered histogram is everything larger than the next to last? If so option 3 would be consistent with that and I think would be preferable on the whole.

Comment by Kevin Arhelger [ 30/Apr/20 ]

I am fine with either 2 (updated CR) or 3. 3 might be slightly less confusing for new users of FTDC.

bruce.lucas do you have a strong opinion either way?

Comment by Mark Benvenuto [ 29/Apr/20 ]

kevin.arhelger and bruce.lucas, here some possible choices for how the buckets will get logged in FTDC when the flag is enabled.

  1. All the buckets (not the preferred choice, original CR)
  2. Only the buckets <= slowMs (updated CR)
  3. Only the buckets <= slowMs + one bucket for > slowMS. This bucket would be labeled int(slowMS+1)
  4. Only the buckets >= slowMs
  5. Some other choice.

Let me know what kind of filtering you would like.

Comment by Bruce Lucas (Inactive) [ 27/Apr/20 ]

kevin.arhelger your ask on this ticket was for histogram buckets <= slowms, but the current code review adds all of the histogram buckets. I think this will have a substantially greater impact on ftdc than your initial ask. Do you think this is still workable from a perspective of retention and impact on consuming tools?

Generated at Thu Feb 08 05:14:49 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.