Prepare chop for production
The settings and surroundings to check before chop serves real traffic, limits, allowed widths, a cache, a read-only container and the startup warnings.
Set the limits, allow the widths you use, and put a cache in front. A server with the defaults accepts originals up to 50 MiB and any output width up to 8192, which is more than a typical site needs. This run lowers the limits, allows four widths and prints an access log:
CHOP_ALLOWED_WIDTHS=320,640,960,1280 \
CHOP_MAX_INPUT_BYTES=10485760 \
CHOP_MAX_INPUT_PIXELS=50000000 \
CHOP_AUTO_FORMATS=webp \
CHOP_ACCESS_LOG=1 \
chop --data /var/www/uploads
An original over the limit is then a 413 with a plain message, and the client learns nothing about your files:
HTTP/1.1 413 Payload Too Large
content-type: text/plain; charset=utf-8
content-length: 32
cache-control: no-store
x-content-type-options: nosniff
The original image is too large.
That output came from CHOP_MAX_INPUT_BYTES=1000 and a 1.3 MB test image.
Checklist
- Cache in front. Responses are built on request and are not stored. Run nginx or a CDN so each variant is built once, and let only the cache reach chop.
- Collapse concurrent misses. Without a lock, every request for an uncached variant makes chop build it. The nginx recipe sets the lock and its timeouts, serves stale copies while one request refreshes them, and shows how to warm a cleared cache. See Avoid stampedes.
- Limit the widths.
CHOP_ALLOWED_WIDTHSbounds the number of variants that anyone can make. See Serve responsive images. - Lower the input limits.
CHOP_MAX_INPUT_BYTES(default52428800) andCHOP_MAX_INPUT_PIXELS(default120000000) bound what one request can make chop decode. Set them to the largest original that you serve. - Size the concurrency.
CHOP_MAX_CONCURRENCYimages are processed at once, by default the number of CPUs, andCHOP_MAX_QUEUE(default64) more wait. A full queue is a503withRetry-After: 1. For a remote origin, the memory in use can reachCHOP_ORIGIN_CONCURRENCY(default4) timesCHOP_MAX_INPUT_BYTES, so lower one of them on a small machine. - Decide on AVIF. AVIF takes more CPU to encode than WebP.
CHOP_AUTO_FORMATS=webpavoids it, and an explicitformat=avifstill works. See Serve AVIF and WebP. - Run read-only. chop never writes to disk, so a container can run with
--read-only --cap-drop ALL. See Run chop in Docker. - Pin the version. Use a version tag of the image or of the npm package, and not
latest. - Check the startup warnings. Each risky setting prints a
warning:line on the standard error:CHOP_ALLOWED_HOSTS=*,CHOP_ORIGIN_ALLOW_HTTP,CHOP_ORIGIN_ALLOW_PRIVATEand an S3 origin without credentials. Fix them or accept them on purpose. - Watch the origin failures. The reason for a refused or failed origin request goes to the standard error as
origin <code>: <reason>, and never to the client. See Troubleshoot chop. - Know that it works.
CHOP_ACCESS_LOG=1prints one JSON line per request with the status and duration, and theX-Cache-Statusheader of the nginx recipe shows hits and misses, including on errors. Watch the502,503and504counts. - Health checks. Use
/healthz. It answersokwithout touching the data directory or the origin. - Authentication and rate limits. chop has neither. Put them in the proxy in front.