Skip to content

[HOLD for payment 2024-09-23] Performance Improvement: Update formatjs libraries and upstream/patch date-fns-tz #47988

Closed

Description

@hurali97

Problem

While we were analysing the latest traces from Jason, we realised that there's substantial amount of time spent in formatjs and date-fns-tz. We also double-checked this with our heavy accounts (~15k reports) and we got the same results. Following is how things look like on baseline:

Baseline

Screenshot 2024-08-26 at 12 47 42 PM

snapshot from profile

Screenshot 2024-08-26 at 12 41 31 PM

Solution:

While analysing the traces, we curated a list of places that can be improved in formatjs and date-fns-tz. Two noticeable items are BestFitMatcher from formatjs and getDateTimeFormat from date-fns-tz. Now, since these are third party libraries, we can bump them to the latest versions and see if they are already fixing performance related issues. We compared the code for BestFitMatcher from the currently installed version formatjs/intl-locale@3.3.0 with formatjs/intl-locale@4.0.0 and there was a significant refactor of this file.

So it only made sense to bump formatjs libraries to the latest to get the performance improvements out of the box. For date-fns-tz there was no significant updates to getDateTimeFormat function, so to apply the improvements we have two options:

  • Patch date-fns-tz
  • Upstream date-fns-tz by creating a PR to it's repository

However, we don't know how long it would take for option 2 as it depends on the reviewer of that repo. We can maybe move forward by mixing both options and removing the patch, once our PR is merged and available on NPM.

The improvement for getDateTImeFormat is pretty simple. We have Intl.DateTimeFormat being re-initialised every time getDateTimeFormat is invoked. Since testDateFormatted is a static variable, we can move it outside of the function.

var dtfCache = {};

// Moved this out from `getDateTimeFormat`
    var testDateFormatted = new Intl.DateTimeFormat('en-US', {
      hour12: false,
      timeZone: 'America/New_York',
      year: 'numeric',
      month: 'numeric',
      day: '2-digit',
      hour: '2-digit',
      minute: '2-digit',
      second: '2-digit'
    }).format(new Date('2014-06-25T04:00:00.123Z'));
    var hourCycleSupported = testDateFormatted === '06/25/2014, 00:00:00' || testDateFormatted === '‎06‎/‎25‎/‎2014‎ ‎00‎:‎00‎:‎00';
    

function getDateTimeFormat(timeZone) {
 ...someImplementation
  return dtfCache[timeZone];
}

Below is how things look like after formatjs version bumps and date-fns-tz patch:

Screenshot 2024-08-26 at 1 03 06 PM

snapshot from profile

Screenshot 2024-08-26 at 12 42 15 PM


To summarise;

  • Bump formatjs libraries to the latest
  • Patch or Upstream date-fns-tz with the changes described above

Activity

  1. melvin-bot commented on Aug 26, 2024

    @melvin-bot
  2. flyinghead commented on Aug 26, 2024

    @flyinghead
  3. melvin-bot commented on Aug 26, 2024

    @melvin-bot
  4. self-assigned this
    on Aug 26, 2024
  5. added
    BugSomething is broken. Auto assigns a BugZero manager.
    on Aug 26, 2024
  6. melvin-bot commented on Aug 26, 2024

    @melvin-bot

    Current assignee @mountiny is eligible for the AutoAssignerNewDotQuality assigner, not assigning anyone new.

  7. melvin-bot commented on Aug 26, 2024

    @melvin-bot

    Triggered auto assignment to @strepanier03 (Bug), see https://stackoverflow.com/c/expensify/questions/14418 for more details. Please add this bug to a GH project, as outlined in the SO.

  8. deleted a comment from on Aug 26, 2024
  9. roryabraham commented on Aug 26, 2024

    @roryabraham
    Contributor

    we can maybe move forward by mixing both options and removing the patch, once our PR is merged and available on NPM

    This is generally my preferred approach. Report the issue with a minimal reproduction, and create a PR to fix it upstream. If you don't receive any feedback, proceed with a patch and create an issue to eventually remove the patch once the upstream fix is merged. If you do get feedback and the maintainer doesn't want to merge your fix for some reason, re-evaluate and open up a conversation with the team.

  10. roryabraham commented on Aug 26, 2024

    @roryabraham
    Contributor

    looks like some good finds @hurali97 👍🏼

  11. 14 remaining items

  12. mountiny commented on Sep 22, 2024

    @mountiny
    Contributor

    @Ollyws sorry, the automation failed here.

    $250 to @Ollyws for their review, no regression tests required for this task cc @strepanier03

  13. mountiny commented on Sep 22, 2024

    @mountiny
    Contributor

    Noting we do not want to close this but make it monthly so we can track of the upstream PR

  14. Ollyws commented on Sep 22, 2024

    @Ollyws
    Contributor

    Requested in ND.

  15. melvin-bot commented on Sep 23, 2024

    @melvin-bot

    Skipping the payment summary for this issue since all the assignees are employees or vendors. If this is incorrect, please manually add the payment summary SO.

  16. garrettmknight commented on Sep 24, 2024

    @garrettmknight
    Contributor

    $250 approved for @Ollyws

  17. mountiny commented on Oct 1, 2024

    @mountiny
    Contributor

    I think we can close now then

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

AutoAssignerNewDotQualityUsed to assign quality issues to engineersAwaiting PaymentAuto-added when associated PR is deployed to productionBugSomething is broken. Auto assigns a BugZero manager.DailyKSv2

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions