Skip to content

feat: add router_rebuild_min_interval to reduce CPU spikes from frequ…#13073

Closed
membphis wants to merge 1 commit into
masterfrom
feat/router-min-rebuild-interval
Closed

feat: add router_rebuild_min_interval to reduce CPU spikes from frequ…#13073
membphis wants to merge 1 commit into
masterfrom
feat/router-min-rebuild-interval

Conversation

@membphis

@membphis membphis commented Mar 5, 2026

Copy link
Copy Markdown
Member

Description

When routes change frequently (e.g., 20-50 changes/sec with 5000+ total routes), the data plane rebuilds the radixtree router on every single change, causing significant CPU spikes (up to 80%) and temporary TPS drops.

This PR adds a configurable router_rebuild_min_interval option that sets a minimum time interval (in seconds) between HTTP router rebuilds. When the interval has not elapsed since the last rebuild, subsequent route changes are acknowledged but the router rebuild is deferred, dramatically reducing CPU usage during bulk route updates.

Configuration

apisix:
  router:
    http: radixtree_host_uri
    ssl: radixtree_sni
    router_rebuild_min_interval: 0  # default: 0 (rebuild immediately, current behavior)

Expected Impact

Scenario (30s duration) min_interval: 0 min_interval: 1 min_interval: 5
20 changes/sec 600 rebuilds ~31 rebuilds ~7 rebuilds
50 changes/sec 1500 rebuilds ~31 rebuilds ~7 rebuilds
1 change/3sec 10 rebuilds 10 rebuilds 10 rebuilds

Changes

  • conf/config.yaml.example: Added router_rebuild_min_interval config option
  • apisix/cli/config.lua: Default value 0
  • apisix/router.lua: Reads config and passes to HTTP router modules (+3 lines)
  • apisix/http/router/radixtree_uri.lua: Min interval check before rebuild
  • apisix/http/router/radixtree_host_uri.lua: Min interval check before rebuild
  • apisix/http/router/radixtree_uri_with_parameter.lua: Min interval check before rebuild
  • t/router/router-rebuild-min-interval.t: Test cases covering default behavior, skip behavior, and deferred rebuild

Backward Compatibility

Default value is 0, preserving the existing behavior of rebuilding immediately on every route change. No existing configurations are affected.

…ent route changes

When routes change frequently (e.g., 20-50 changes/sec), the router is
rebuilt on every change which causes CPU spikes under high-frequency
updates due to full radixtree recompilation with all routes.

This adds a configurable minimum interval between router rebuilds.
When set to a positive value (e.g., 1 or 5 seconds), rebuilds are
throttled so the router is not recompiled more frequently than the
configured interval, significantly reducing CPU usage during bulk
route updates.

Default value is 0 (rebuild immediately on every change), preserving
backward compatibility with existing behavior.
@moonming

Copy link
Copy Markdown
Member

Architectural Note (self-review):

This PR (#13073) and #12826 (incremental router compilation by @apaloleg) address the same problem — CPU spikes from frequent router rebuilds — with different approaches:

  • This PR: router_rebuild_min_interval — throttle/debounce rebuilds (quick win, low risk)
  • feat: incremental compilation routers #12826: Incremental compilation — only rebuild affected parts of the radix tree (fundamental fix, higher complexity)

Proposed path forward:

  1. Merge this throttling approach first as an immediate relief (low risk, easy to understand)
  2. Continue developing incremental compilation as a longer-term optimization
  3. The two approaches are complementary — throttling reduces frequency, incremental reduces cost per rebuild

Will coordinate with @apaloleg on #12826 to align timelines.

@github-actions

Copy link
Copy Markdown

This pull request has been marked as stale due to 60 days of inactivity. It will be closed in 4 weeks if no further activity occurs. If you think that's incorrect or this pull request should instead be reviewed, please simply write any comment. Even if closed, you can still revive the PR at any time or discuss it on the dev@apisix.apache.org list. Thank you for your contributions.

@github-actions github-actions Bot added the stale label May 16, 2026
@github-actions

Copy link
Copy Markdown

This pull request/issue has been closed due to lack of activity. If you think that is incorrect, or the pull request requires review, you can revive the PR at any time.

@github-actions github-actions Bot closed this Jun 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants