DNS histograms rewrite |
|||
Issue descriptionThe async and system resolvers each have their own set of UMA histograms measuring mostly, but not exactly, the same things. This structure works against the Finch tools for experiments, which compare the same histogram across different experiment groups, and more generally makes it hard to compare the implementations. Before running an experiment, the histograms need to be rewritten. Doc: https://docs.google.com/a/google.com/document/d/1X1wsXwmMdwVa15AXMOdN8Z61gaKbb0cnntbUb61YdR8/edit?usp=sharing
,
Sep 28 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/90d05e08a3f97289971e1d54f488287a42e3942b commit 90d05e08a3f97289971e1d54f488287a42e3942b Author: Miriam Gershenson <mgersh@chromium.org> Date: Thu Sep 28 19:29:28 2017 DNS histogram cleanup, part 2 This CL merges the higher-level DNS histograms to include both async and system resolver timing in the same metrics. This will allow easier use in Finch experiments. It also adds the new Net.DNS.TotalTimeNotCached histogram, which is equivalent to Net.DNS.TotalTime minus cached and other synchronous resolutions, for better alerts. The current TotalTime histogram is hard to use for alerts because the median is 0. Bug: 769320 Change-Id: Iea7873ddd284fffe1e0e414f1e33f2e07053c0ca Reviewed-on: https://chromium-review.googlesource.com/673092 Commit-Queue: Miriam Gershenson <mgersh@chromium.org> Reviewed-by: Julia Tuttle <juliatuttle@chromium.org> Reviewed-by: Ilya Sherman <isherman@chromium.org> Cr-Commit-Position: refs/heads/master@{#505119} [modify] https://crrev.com/90d05e08a3f97289971e1d54f488287a42e3942b/net/dns/host_resolver_impl.cc [modify] https://crrev.com/90d05e08a3f97289971e1d54f488287a42e3942b/tools/metrics/histograms/histograms.xml
,
Sep 28 2017
,
Sep 28 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/6160466523bb411e66ab80a880ff05a248381d9c commit 6160466523bb411e66ab80a880ff05a248381d9c Author: Miriam Gershenson <mgersh@chromium.org> Date: Thu Sep 28 23:51:11 2017 DNS histogram cleanup, part 3 This CL removes all the old DNS.Resolve* and AsyncDNS.Resolve* histograms and replaces them with a new set. The new Net.DNS.Resolve* histograms measure timing and success/failure at the Job layer, and each type of Task gets its own set to measure Task layer timing and success/failure. Bug: 769320 Change-Id: I0b0df897310a474fd11022f932724c8d40789090 Reviewed-on: https://chromium-review.googlesource.com/687836 Commit-Queue: Miriam Gershenson <mgersh@chromium.org> Reviewed-by: Julia Tuttle <juliatuttle@chromium.org> Reviewed-by: Ilya Sherman <isherman@chromium.org> Cr-Commit-Position: refs/heads/master@{#505215} [modify] https://crrev.com/6160466523bb411e66ab80a880ff05a248381d9c/net/dns/host_resolver_impl.cc [modify] https://crrev.com/6160466523bb411e66ab80a880ff05a248381d9c/tools/metrics/histograms/histograms.xml
,
Nov 27 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/aaf139580bc8d4e6816e358ea22674967485bc30 commit aaf139580bc8d4e6816e358ea22674967485bc30 Author: Miriam Gershenson <mgersh@chromium.org> Date: Mon Nov 27 16:06:06 2017 Add histograms for DNS error codes DNS failure is very common, and there is currently no tracking of overall failure causes. Histograms are separated by how fast the failure happened because there's a large clump of almost immediate failures that I want to investigate further, and causes are probably different between the two categories. 10 ms was chosen as the cutoff point by looking at the Net.DNS.ResolveFailureTime histogram. Bug: 769320 Change-Id: I15fa7105d2208d47799935141faf3b118b52e0b2 Reviewed-on: https://chromium-review.googlesource.com/782662 Reviewed-by: Steven Holte <holte@chromium.org> Reviewed-by: Julia Tuttle <juliatuttle@chromium.org> Commit-Queue: Miriam Gershenson <mgersh@chromium.org> Cr-Commit-Position: refs/heads/master@{#519312} [modify] https://crrev.com/aaf139580bc8d4e6816e358ea22674967485bc30/net/dns/host_resolver_impl.cc [modify] https://crrev.com/aaf139580bc8d4e6816e358ea22674967485bc30/tools/metrics/histograms/histograms.xml
,
Feb 7 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/19faef8176944732b29b2e2c2d2c25f594a8c3d4 commit 19faef8176944732b29b2e2c2d2c25f594a8c3d4 Author: Miriam Gershenson <mgersh@chromium.org> Date: Wed Feb 07 23:56:44 2018 Add histogram to track all DnsTask errors There is already a histogram to track DnsTask errors that are followed by a ProcTask success, and a pair of histograms that track overall error codes for the host resolver. Overall error codes always come from the system resolver because of the fallback, and these error codes aren't very descriptive. Tracking DnsTask errors will give a better picture of error causes. Bug: 769320 Change-Id: I40259169417168031cf31dda1c65d48e5636c4f7 Reviewed-on: https://chromium-review.googlesource.com/905262 Reviewed-by: Steven Holte <holte@chromium.org> Reviewed-by: Julia Tuttle <juliatuttle@chromium.org> Commit-Queue: Miriam Gershenson <mgersh@chromium.org> Cr-Commit-Position: refs/heads/master@{#535195} [modify] https://crrev.com/19faef8176944732b29b2e2c2d2c25f594a8c3d4/net/dns/host_resolver_impl.cc [modify] https://crrev.com/19faef8176944732b29b2e2c2d2c25f594a8c3d4/tools/metrics/histograms/histograms.xml
,
Mar 2 2018
I didn't do everything in the original doc, but I think I'm going to leave it as it is now because the rest is just unnecessary churn. Marking this fixed. |
|||
►
Sign in to add a comment |
|||
Comment 1 by bugdroid1@chromium.org
, Sep 28 2017