Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Demand control lookup optimizations #6450

Open
wants to merge 12 commits into
base: dev
Choose a base branch
from

Conversation

tninesling
Copy link
Contributor

@tninesling tninesling commented Dec 12, 2024

Overview

Demand Control can reduce router throughput due to the extra processing required for scoring. Currently, our simple load test suite indicates somewhere in the range of 10-15% throughput reduction, with much of the time spent in data lookups. This change shifts more data to be computed at plugin initialization and consolidates lookup queries.

  • Cost directives for arguments are now stored in a map alongside those for field definitions
  • All precomputed directives are bundled into a struct for each field, along with that field's extended schema type. This reduces 5 individual lookups into 1
  • Response visitor only performs one lookup per visited field, where it was doing so twice previously.

I do think there is still an opportunity to cut the number of lookups roughly in half by removing indirection. Currently, the precomputed directives are in a two-layer map which mirrors apollo_compiler::Schema::type_field(). Ideally, we could get each child field's definition from its parent as we traverse down the document instead of looking up by type and field name, but I haven't been able to encode that potentially recursive structure in a way that pleases the compiler.

Perf Results

These results reference the following data:
Baseline: 9679d68 flamegraph
Target: 1140891 flamegraph

Overall, the demand-control-instrumented simulation shows that demand control CPU time has been decreased from 7.5% to 6.1% for that scenario.

Request Scoring

Before:
Screenshot 2024-12-13 at 12 13 05
After:
Screenshot 2024-12-13 at 12 14 46

Response Scoring

Before:
Screenshot 2024-12-13 at 12 15 13
After:
Screenshot 2024-12-13 at 12 15 35


Checklist

Complete the checklist (and note appropriate exceptions) before the PR is marked ready-for-review.

  • Changes are compatible1
  • Documentation2 completed
  • Performance impact assessed and acceptable
  • Tests added and passing3
    • Unit Tests
    • Integration Tests
    • Manual Tests

Exceptions

Note any exceptions here

Notes

Footnotes

  1. It may be appropriate to bring upcoming changes to the attention of other (impacted) groups. Please endeavour to do this before seeking PR approval. The mechanism for doing this will vary considerably, so use your judgement as to how and when to do this.

  2. Configuration is an important part of many changes. Where applicable please try to document configuration examples.

  3. Tick whichever testing boxes are applicable. If you are adding Manual Tests, please document the manual testing (extensively) in the Exceptions.

@svc-apollo-docs
Copy link
Collaborator

svc-apollo-docs commented Dec 12, 2024

✅ Docs Preview Ready

No new or changed pages found.

This comment has been minimized.

@router-perf
Copy link

router-perf bot commented Dec 12, 2024

CI performance tests

  • connectors-const - Connectors stress test that runs with a constant number of users
  • const - Basic stress test that runs with a constant number of users
  • demand-control-instrumented - A copy of the step test, but with demand control monitoring and metrics enabled
  • demand-control-uninstrumented - A copy of the step test, but with demand control monitoring enabled
  • enhanced-signature - Enhanced signature enabled
  • events - Stress test for events with a lot of users and deduplication ENABLED
  • events_big_cap_high_rate - Stress test for events with a lot of users, deduplication enabled and high rate event with a big queue capacity
  • events_big_cap_high_rate_callback - Stress test for events with a lot of users, deduplication enabled and high rate event with a big queue capacity using callback mode
  • events_callback - Stress test for events with a lot of users and deduplication ENABLED in callback mode
  • events_without_dedup - Stress test for events with a lot of users and deduplication DISABLED
  • events_without_dedup_callback - Stress test for events with a lot of users and deduplication DISABLED using callback mode
  • extended-reference-mode - Extended reference mode enabled
  • large-request - Stress test with a 1 MB request payload
  • no-tracing - Basic stress test, no tracing
  • reload - Reload test over a long period of time at a constant rate of users
  • step-jemalloc-tuning - Clone of the basic stress test for jemalloc tuning
  • step-local-metrics - Field stats that are generated from the router rather than FTV1
  • step-with-prometheus - A copy of the step test with the Prometheus metrics exporter enabled
  • step - Basic stress test that steps up the number of users over time
  • xlarge-request - Stress test with 10 MB request payload
  • xxlarge-request - Stress test with 100 MB request payload

@tninesling tninesling requested a review from BrynCooke December 13, 2024 00:03
@tninesling tninesling marked this pull request as ready for review December 13, 2024 00:03
@tninesling tninesling requested review from a team as code owners December 13, 2024 00:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants