Performance Triage

From MozillaWiki
Jump to: navigation, search


Idea.png
If you have any feedback/suggestions/questions regarding the performance triage process, you can share them in #perf-triage, or reach out to Dave Hunt or Frank Doty.

Nomination

Bugzilla

To (re)nominate a bug for triage, set the Performance Impact flag in Bugzilla to ?

This can be found by clicking Show Advanced Fields followed by Set bug flags when entering a new bug:

Bugzilla performance nomination on new bug form.png

Or by expanding the Tracking section when editing an existing bug:

Screenshot 2022-02-24 at 19.53.54.png

GitHub

To nominate a bug for triage, add the Performance label to an issue. This can be done by filing an new issue with the "Performance issue" template:

Screenshot of file a "Performance issue" template on GitHub

Or by opening an existing issue on GitHub and selecting the label from the right-hand bar:

Screenshot of adding a performance label on GitHub

Currently, only the following GitHub repositories are supported:

Queries

Performance triage

Full Query
ID Summary Status
1419502 Specific PDF page takes 10s to render (200% cpu) REOPENED
1436734 pdf with many images uses a lot of memory on scrolling NEW
1445735 Unable to render specific PDF file. NEW
1485065 Firefox Pdf single page document loads several minutes while other browsers instantly (attached sample PDF) NEW
1536420 PDF uses lot of memory when decoding UNCONFIRMED
1731514 Extreme memory surge when opening a certain pdf-file NEW
1785979 Slow rendering and opening of certain PDF types NEW
1874013 Incorrect and slow rendering of some PDF files NEW
1878133 PDF takes excessively long to render NEW
1887068 Add support for perfetto track events NEW
1888064 Firefox takes a long time to start prior to Load a Website UNCONFIRMED
1888322 Latest update crashes and cause processor loop crash. UNCONFIRMED
1888337 Significantly worsened keyboard auto-repeat performance in Fx 124.0.1 NEW
1888525 helpdeskgeek site runs a lot more JS in Firefox than in Chrome (checkerboards during scrolling) NEW

14 Total; 14 Open (100%); 0 Resolved (0%); 0 Verified (0%);


Performance triage (pending-needinfo)

Full Query
ID Summary Status
1805255 Firefox has 50% more soft page faults than Chrome running Speedometer 2 NEW
1866145 Delay opening new windows due to long GC slices in the parent process NEW
1870209 Webpage sometimes no response in Fenix with high CPU usage UNCONFIRMED
1870532 Firefox 120.0.1 grabs 100% of cpu resources and freezes UNCONFIRMED
1872368 Since the last update 121.0 Firefox uses insane amount of Ram 99% of my 32GB without any real reason and if it gets closed it buges out for 10 seconds spikes over the ram limit just to slowly close UNCONFIRMED
1872414 Firefox lags after latest update, high memory use (121.0 12-2023) UNCONFIRMED
1875547 Slow (single-threaded) PDF rendering for documents with lots of vector shapes NEW
1876708 firefox using 3.4gb of memory with a few tabs open UNCONFIRMED
1877007 extremely high CPU usage even with one tab open on the splash screen UNCONFIRMED
1879177 3 youtube tabs are using 7GB of RAM UNCONFIRMED
1879905 firefox consuming 2 GB of RAM with 8 tabs open UNCONFIRMED
1882916 Firefox consumes a lot of CPU and makes me Mac M2 hang for a good 15 seconds UNCONFIRMED
1882952 Firefox Browser Freezes when Opening Multiple Tabs UNCONFIRMED
1883246 firefox using large amount of memory UNCONFIRMED
1883475 huge amount of memory storage (constantly) idk why pls help/................................................................................................................................................. UNCONFIRMED
1883507 Memory usage is too high 23GB and up to 60% CPU usage. UNCONFIRMED
1883669 Memory Problem UNCONFIRMED
1884296 Firefox is using lots of memory when only having several tabs open UNCONFIRMED
1885796 Use to many Ram UNCONFIRMED

19 Total; 19 Open (100%); 0 Resolved (0%); 0 Verified (0%);


Recently opened bugs with performance keywords in the summary

Full Query
ID Summary Status
1885627 [Menu Redesign] Implement Translation states NEW
1885657 Clicking on “always translate <code> language” doesn’t Immediately Translate NEW
1885661 Sharp Cut-off in Translations Language Dropdown NEW
1885662 Remove capitalization in "Translation Options", should be "Translation options" NEW
1885667 Update Translations Padding NEW
1885672 Match Android UI Toolbar Color in Translations NEW
1885677 Translations Error State Refinements NEW
1886079 Extremely slow animations and scrolling performance on B2B2go's website UNCONFIRMED
1886221 Add Telemetry to send "to" and "from" languages for each Translation NEW
1886243 Adjust translation "Translate" button color NEW
1886298 consider changing `autofill to the next slash` to avoid generating invalid URLs NEW
1886335 [WebGPU] Demo at https://codepen.io/dawken/pen/XWPNJBx doesnt draw anything in Nightly NEW
1886336 [WebGPU] Demo at https://codepen.io/eriklu/pen/zYMeMGQ takes a long time to run (and never displays anything on the screen) NEW
1886429 Hide "Translation preferences"purple dialog header if Nimbus isn't controlling any of those areas NEW
1886699 Crash Annotation GraphicsCriticalError: |[0][GFX1-]: Killing GPU process due to IPC reply timeout (t=86464.5) [GFX1-]: Killing GPU process due to IPC reply timeout UNCONFIRMED
1886753 Broken context menu in sidebar due to missing translations fluent reference NEW
1886780 5.61 - 5.19% perf_reftest_singletons many-custom-props.html / perf_reftest_singletons many-custom-props.html (OSX) regression on Mon March 11 2024 NEW
1886851 Change Nimbus Exposure Event and Add Telemetry when an Expected or Offer Translation Event Occurs NEW
1886955 SVG animateMotion not using GPU? UNCONFIRMED
1886982 Number.toFixed() is slow NEW
1887090 Rename about:translations translations.mjs to aboutTranslations.mjs NEW
1887140 Accesibility Review of Translations Main Flow, Translation Page Settings, and First Tier of Translation Global Settings for Android ASSIGNED
1887287 Closing webgpu demos doesnt release memory, making testing webgpu on Nightly very difficult NEW
1887300 Codepen demo (https://codepen.io/internette/pen/yqpQpE) is ~2x slower in Nightly NEW
1887312 ONNX benchmark for WASM(INT8) is 2x-3x slower in Nightly NEW
1887423 Failing WPT css/css-transforms/translate/translate-and-transform-css-property-in-svg.html NEW
1887519 [WebGPU] A minimized tab with webgpu will have instant multi-GB memory increase on maximizing NEW
1887529 "More Translation Settings" expanding for a split second when opening about:preferences for the first time within the same session ASSIGNED
1887554 Video re-encode demo has rapid unbound system-level memory growth. (Chrome doesnt do anything there) NEW
1887557 about:logging preset for WebGPU doesnt not create any output to a log file (logging to the profiler works) NEW
1887574 Move bundle recording wgpu-core calls into `gfx/wgpu_bindings` NEW
1887582 Modified Codepen demo is faster to load in Nightly but uses tons of more RAM NEW
1887811 my firefox browser are crashing and don't open any thing, it's very slow UNCONFIRMED
1887864 The variant `media-gpu` expiration is on 2024-04-21 NEW
1887909 Update WGPU to upstream (week of 2024-03-26) ASSIGNED
1887920 Todoist triggers high CPU via constantly painting infinite SVG animations NEW
1887945 Review Enabled and Disabled Buttons in Translation States NEW
1887950 Attribute GPU process memory to a particular content process NEW
1887953 Improve SelectTranslationsPanel language detection algorithm NEW
1887997 URL.searchParams performance issue NEW
1888011 The heading of the Translations pop-up sheet is not announced as such by TalkBack NEW
1888015 The list of the Translations settings is missing the length and position information in the TalkBack announcement NEW
1888109 Can new Worker() be made to properly operate on the background (maybe when the URL is an in-memory Blob)? NEW
1888174 Hit MOZ_CRASH(Cannot remove a vacant resource) at /builds/worker/checkouts/gecko/third_party/rust/wgpu-core/src/storage.rs:207 NEW
1888175 Hit MOZ_CRASH(Texture[Id(1,1,d3d12)] does not exist) at /builds/worker/checkouts/gecko/third_party/rust/wgpu-core/src/storage.rs:110 NEW
1888176 Assertion failure: false (MOZ_ASSERT_UNREACHABLE: unexpected to be called), at /builds/worker/checkouts/gecko/dom/webgpu/ipc/WebGPUParent.cpp:1535 NEW
1888178 Hit MOZ_CRASH(index out of bounds: the len is 2 but the index is 2) at /builds/worker/checkouts/gecko/third_party/rust/wgpu-core/src/storage.rs:201 NEW
1888326 SplitRun() doesn't advance the index after committing a region of memory ASSIGNED
1888328 Add a profiler marker for purging memory from mozjemalloc NEW
1888354 poor Youtube video performance when browser stays open for long time UNCONFIRMED
1888439 Memory leak on Business Insider UNCONFIRMED
1888514 Improve edge case error handling in Translations setting pages NEW
1888588 Rename MaybeGetSurfaceDescriptorForCanvasTranslator() to MaybeGetSurfaceDescriptorForRemoteCanvas() NEW
1888610 [GPU-canvas] A simple canvas demo stutters when opened in two foreground tabs (without using much GPU) NEW

54 Total; 54 Open (100%); 0 Resolved (0%); 0 Verified (0%);


Triage process

Introduction

The goal of performance triage is to identify the extent to which bugs impact the performance of our products, and to move these bugs towards an actionable state. The goal is not to diagnose or fix bugs during triage. We triage bugs that have been nominated for triage and bugs in the Core::Performance component that do not have the performance impact project flag set.

During triage we may do any/all of the following:

  • Request further information from the reporter (such as a profile)
  • Set the performance impact project flag
  • Add performance keywords
  • Move the bug to a more appropriate component

Who is responsible for triage?

Everyone is welcome to take part in triage. By default, everyone on the performance team is enrolled in triage rotation, but we also have participants from outside the team.

How do I schedule a triage meeting?

If you are on triage duty, you will receive an invitation as a reminder to schedule the triage meeting on the shared performance calendar with the nominated sheriffs invited at a time that works for them. The responsibility of scheduling the meeting falls to the lead sheriff. Once a triage meeting has been scheduled, it’s a good idea to remove the reminder event from the calendar to avoid confusion. It’s a good idea to use the shared calendar, as this increases the visibility of the performance triage and allows other members of the team to contribute or observe the process.

What if a sheriff is unavailable?

The rotation script is not perfect, and doesn’t know when people are on PTO or otherwise unavailable. If the lead sheriff is available, it is their responsibility to either schedule the triage with the remaining available sheriff or to identify a suitable substitute for the unavailable sheriff(s). If the lead sheriff is unavailable, this responsibility passes onto the remaining available sheriffs.

How do I run a triage meeting?

The following describes the triage process to follow during the meeting:

  1. Ask if others would prefer you to share your screen. This can be especially helpful for those new to triage.
  2. Open the first triage query to show bugs nominated for triage or in the Core::Performance component without the performance impact project flag set. The bugs are sorted from oldest to newest. For each bug in the list, follow these steps:
    • Bugs that look like tasks that were filed by members of the Performance team will generally need to be moved to the Core::Performance Engineering component.
    • For defects: Determine if the bug is reproducible and actionable. If not, add a needinfo for the reporter asking for more information, set the performance impact project flag to pending-needinfo, and then move onto the next bug. We have a template that you can modify as needed.
    • For all bugs (including enhancements):
  3. Open the second triage query to show bugs that are waiting further information to determine the performance impact. The bugs are sorted from oldest to newest. For each bug in the list, follow these steps:
    • If the performance impact project flag was set to pending-needinfo less than 2 weeks ago, move onto the next bug.
    • If the performance impact project flag was set to pending-needinfo more than 2 weeks ago but less than 2 months ago, consider adding a needinfo for either: another reporter of the issue, someone with access to the appropriate platform(s) to attempt to reproduce the issue, or a relevant subject matter expert.
    • If the performance impact project flag was set to pending-needinfo more than 2 months ago, close the bug as inactive. You can modify the inactive bug template as needed.
  4. If time permits, open the third triage query to show recently opened bugs with performance related keywords in the summary. If any of these look like performance bugs, they can either be triaged the same way as bugs in the initial query or they can be nominated for triage in a subsequent meeting.

What if things don't go as expected?

Don't panic! The triage process is not expected to be perfect, and can improve with your feedback. Maybe the result of the triage calculator doesn't feel right, or you find a scenario that's not covered in these guidelines. In this case we recommend that you bring it up in #perf-triage, or consider scheduling a short meeting with some triage leads (you can see some recent leads in the triage rotation). If in doubt, leave a comment on the bug with your thoughts and move on. There's a chance someone will respond, but if not the next performance triage sheriffs may have some other ideas.

How do I determine the performance impact project flag?

The performance impact project flag is used to indicate a bug’s relationship to the performance of our products. It can be applied to all bugs, and not only defects. The triage calculator should be used to help determine the most appropriate value for this flag. In addition to setting the performance impact project flag, make sure to use the “Copy Bugzilla Comment” button and paste this as a comment on the bug.

If you do not have enough information to set the performance impact project flag, open a needinfo request against an appropriate individual (such as a reporter), and set the performance impact project flag to pending-needinfo.

For more information about what this flag, and it's settings mean see this blog post.

How do I determine the performance keywords?

There are several performance related keywords, which can be helpful to understand how our performance issues are distributed, or whenever there’s a concerted effort to improve a particular aspect of our products. The triage calculator may recommend keywords to set, and by typing “perf:” in the keywords field in Bugzilla, you will see the available options. Select all that apply to the bug.

How do I determine the correct Bugzilla component?

Ideally we would only have bugs in the Core::Performance component that are the responsibility of the engineers in the performance team. For performance bugs to have the best chance of being fixed, it's important to assign them to the correct component. In some cases the correct component will be obvious from the bug summary, description, or steps to reproduce. In other cases, you may need to do a bit more work to identify the component. For example, if there's a profile associated with the bug, you could see where the majority of time is being spent using the category annotations.

How do I read a performance profile?

It's useful to be able to understand a profile generated by the Firefox Profiler, and hopefully someone in the triage meeting will be able to help. If you find an interesting profile, or just want to understand how to use them to analyse a performance problem, we encourage you to post a link to the profile (or bug) in #joy-of-profiling where someone will be happy to help. The profile may even be analysed during one of the regular "Joy of Profiling" open sessions that can be found on the Performance Office Hours calendar.

Triage calculator

The Performance Impact Calculator was developed to assist in identifying and applying the performance impact project flag and performance keywords consistently. If you have feedback or would like to suggest changes to this tool, please share these in the #perf-triage Matrix channel.

Triage rotation

The sheriffs are allocated on a weekly basis, which is published here. The rotation is generated by this script.

Templates

New bug

This template is included in the description for new bugs opened in the Core::Performance component. If a bug is opened in another component and then moved to Core::Performance, this template can be used as needed to request additional information from the reporter.

### Basic information

Steps to Reproduce:


Expected Results:


Actual Results:


---

### Performance recording (profile)

Profile URL:
(If this report is about slow performance or high CPU usage, please capture a performance profile by following the instructions at https://profiler.firefox.com/. Then upload the profile and insert the link here.)

#### System configuration:

OS version:
GPU model:
Number of cores: 
Amount of memory (RAM): 

### More information

Please consider attaching the following information after filing this bug, if relevant:

 - Screenshot / screen recording
 - Anonymized about:memory dump, for issues with memory usage
 - Troubleshooting information: Go to about:support, click "Copy text to clipboard", paste it to a file, save it, and attach the file here.

---

Thanks so much for your help.

Moved to Core::Performance

This bug was moved into the Performance component. Reporter, could you make sure the following information is on this bug?

 - For slowness or high CPU usage, capture a profile with http://profiler.firefox.com/ , upload it and share the link here.
 - For memory usage issues, capture a memory dump from about:memory and attach it to this bug.
 - Troubleshooting information: Go to about:support, click "Copy raw data to clipboard", paste it into a file, save it, and attach the file here.

Thank you.

No longer able to reproduce

This bug doesn’t seem to happen anymore in current versions of Firefox. Please reopen or file a new bug if you see it again.

No response from reporter

With no answer from the reporter, we don’t have enough data to reproduce and/or fix this issue. Please reopen or file a new bug with more information if you see it again.

Expected behaviour

This is expected behavior. Please reopen or file a new bug if you think otherwise.

Website issue

According to the investigation, this is a website issue. Please reopen or file a new bug if you think otherwise.