Issue metadata
Sign in to add a comment
|
Regression: compile output now has timestamps |
||||||||||||||||||||||
Issue descriptionThis makes the output much larger, and makes the build output more noisy and way slower to download, and is 100% redundant with the ninja trace that goma uploads and which has an about:tracing style output. (See also lengthy discussion in https://codereview.chromium.org/483853002/ and https://bugs.chromium.org/p/chromium/issues/detail?id=727451#c14 onward.
,
Sep 27
It should default off for compile output, and probably for test output. That output is already large and takes a long time to download.
,
Sep 27
So if I understand correctly, there are two concerns: 1. Clutter 2. Speed Visiting speed: Compare no timestamps: https://3501-b4019a5-dot-luci-logdog.appspot.com/logs/chromium/buildbucket/cr-buildbucket.appspot.com/8934275041283648992/+/steps/compile__with_patch_/0/stdout With timestamps: https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket.appspot.com/8934275041283648992/+/steps/compile__with_patch_/0/stdout No timestamps: 108KB, load time ~700ms With timestamps: 236KB, load time 300ms - 1200ms (varies widely). Most of the time is spent in transport. I believe the bottleneck is most likely fetching entries out of bigtable, which is the same regardless of whether or not the timestamps are there. In this case, I think any slowness attributed to timestamps is mostly eaten by variance in bigtable (or whatever backend0 loading latency. We might see a more pronounced effect on a larger log stream, though. So if we have a meaningful switch or option to turn off timestamps for speed reason, it can't be on the client side, it would have to be a URL param or cookie. Visiting clutter: The timestamps do add a fair amount of visual clutter to the log view. Currently the gutter is 100px wide, and set to the same color as the log lines. If we make the gutter smaller (60px is probably still safe), and fade out the text, I think it would be less clutter. Original: https://screenshot.googleplex.com/iojUHTePms1 Faded: https://screenshot.googleplex.com/Adj26rAHdKs
,
Sep 27
Assuming fixed bandwidth, half the page size should translate into half the load time, no?
,
Sep 27
Can we make this an opt-in thing for whatever view this was requested for?
,
Sep 27
re #4: it is only if we assume the backend could serve logs at the full speed of the network pipe. This isn't quite true, since we're bottlenecked on the rendering speed of the backend. Even without timestamps, the load speed is about 1.2Mb/s, which is less than my 1Gb/s connection. user option is probably the way to go, for this.
,
Sep 27
Looks like the times are also colored: https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket.appspot.com/8934205803885461280/+/steps/compile__with_patch_/0/stdout This is actively misleading for compile output since a bunch of tasks run in parallel and printed when completed. So if something takes a long time but completes right after some fast task, it shows up as fast. That's why we have that about:tracing view of the build already, which displays parallelism.
,
Sep 27
If there's going to be an user-side option, then let's please default to off.
,
Sep 28
hinoka@ and I discussed this today. We should default to timestamps being off (as you say), but being to enable them will be a nice feature generically for logs, even if for some logs it'll either be redundant or actively confusing and hence shouldn't be shown.
,
Sep 28
Also adding in discission with iannucci@: Half of the feature requests for logdog are to add some sort of enhancements to the logs, for example linkification, subset of html tags, color highlighting for error severity, etc. The proposal is to reduce modes have two modes for logdog: lite and featured. Lite will be like the old plain-text mode, and logs are stripped of all features, and the emphasis is on minimum amount of features in exchange for speed. Featured mode will include all features, including timestamps and the other FRs mentioned above. The setting will be stored as a cookie (instead of user-settings, which adds complexity), and we can default to lite mode. This is in addition to the raw mode currently available.
,
Sep 28
Sounds great, thanks!
,
Sep 28
,
Oct 9
This has been around for close to two weeks now. Is it difficult to revert? What's the hold-up?
,
Oct 16
The following revision refers to this bug: https://chromium.googlesource.com/infra/luci/luci-go.git/+/99af80510db303195ab78634251eab774699c968 commit 99af80510db303195ab78634251eab774699c968 Author: Ryan Tseng <hinoka@google.com> Date: Tue Oct 16 17:01:53 2018 [logdog] Add switch for lite mode. Bug: 889947 Change-Id: I8595b6c5b58f7215861454bff88d34e6cfea9255 Reviewed-on: https://chromium-review.googlesource.com/c/1271236 Commit-Queue: Ryan Tseng <hinoka@chromium.org> Reviewed-by: Nodir Turakulov <nodir@chromium.org> [modify] https://crrev.com/99af80510db303195ab78634251eab774699c968/logdog/appengine/cmd/coordinator/static/static/css/viewer.css [modify] https://crrev.com/99af80510db303195ab78634251eab774699c968/logdog/appengine/cmd/coordinator/static/static/js/viewer.js [modify] https://crrev.com/99af80510db303195ab78634251eab774699c968/logdog/appengine/coordinator/flex/logs/http.go
,
Oct 16
Timestamps are now hidden behind a link. Compared to the original implementation, the log in #3 went from 108KB to 95KB.
,
Oct 16
👍 Thanks! |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by hinoka@chromium.org
, Sep 27