DocsLimits & performance

Help

Limits & performance

Everything in Akev that grows with traffic has a limit by default, so a capture that runs all day levels off instead of filling your memory or disk. When a limit is reached, Akev says so on the exchange or in a notice, and Settings → Capture → Limits reached since Akev started counts each kind.

The limits

What Setting (default) When it's reached
Exchanges kept Settings → Capture → Max exchanges (200,000) The oldest exchanges that aren't pinned are removed.
Body size Max body size (32 MB) Larger bodies are cut off at this size; the exchange is still recorded and marked truncated. The client always gets the whole body.
Bodies in memory Memory budget (512 MB) The oldest bodies move to disk, compressed.
Bodies on disk Disk space for bodies (4 GB) The oldest bodies on disk are deleted; their exchanges stay, marked as having lost the body.
Traffic waiting to be stored Capture queue (64 MB) New exchanges wait up to a quarter of a second, then their bodies are dropped, with a notice.
Bodies held for rules Memory for rule bodies (1 GB, shared by all exchanges) A body a rule, script or breakpoint would need in full streams through untouched instead, and the exchange says Body too large to apply rules; passed through.
Exchanges paused at breakpoints Settings → Proxy → Breakpoints → Max paused at once (100) Further matches continue without pausing, and each says why.
Waiting for a response Response timeout (120 s) The exchange fails with a timeout when the server hasn't started answering.
A silent response body Body idle timeout (5 min) The exchange fails with a timeout. Streams that keep sending are never cut off.
Autosave Settings → Data → Keep at most (256 MB; 20,000 exchanges) The newest exchanges are kept in the autosave; see sessions.
Requests to the API 2 MB; 32 MB for rules; 1 GB for imports and Compose The API answers 413.

Each setting is 0 for no limit, and also in settings.json and the settings API (capture.maxExchanges, capture.memoryBudgetBytes, capture.spillMaxBytes, capture.eventQueueBytes, proxy.ruleBodyBudgetBytes, breakpoints.maxPaused, proxy.responseTimeoutMs, proxy.bodyIdleTimeoutMs, sessions.autosaveMaxBytes and sessions.autosaveMaxExchanges). See also timeouts.

With the default limits, a long session of typical API traffic levels off at the size of 200,000 exchanges: about 750–900 MB. Halving Max exchanges keeps it under 400 MB.

Measured performance

What Result
Throughput through the proxy, 64 clients at once, 1 KB responses About 17,600 requests per second for 100,000 requests (p50 2.6 ms, p99 13.9 ms); about 15,700 per second for 200,000 more at the exchange limit; no failures
Memory with 100,000 exchanges captured About 360 MB
Memory when idle About 22 MB
Filtering 100,000 exchanges by URL, through the API 19 ms

Method. A release build of the akev command (without link-time optimisation, so the published builds are a little faster), with an ephemeral data directory and the default settings, on a 2015 desktop (Intel Core i7-6700, 4 cores, 31 GB of RAM, Fedora Linux) that was busy with other builds during every run, so treat throughput and latency as lower bounds. akev-loadgen, in the source's crates/akev-loadgen, sent GET requests with browser-like headers through the proxy to a local server answering with about 1 KB of JSON:

bash
akev --token <token> start --ephemeral --no-open --proxy-port 18000 --api-port 18456
akev-loadgen --proxy http://127.0.0.1:18000 --requests 100000 --concurrency 64 --body-size 1024

Memory is the process's resident set size, without the UI open. Your numbers depend on your hardware and your traffic: run it yourself.

Type to search guides, the CLI and the API.