A live, per-phase HTTP timing probe. Like curl -w on steroids: every phase prints the moment it lands, so a hung request stalls right where it's stuck.
curl -w wishes it did.Client-side, per-phase evidence for the endpoints you don't control. When the owner says "looks fine on our end", point at the column that isn't.
liveEach phase prints as it completes. A hang stalls visibly at the phase that's stuck.
deltasEvery column is that phase alone, not a running total. Only TOTAL_TIME accumulates.
--statsmin, p50 through p99, max, mean and stdev per phase across -c N runs. Tail latency, not averages.
--assert-status --max-ttfbExits non-zero when a status or phase budget is breached. Drops straight into pipelines.
--expect-body --expect-regexFail when the body is wrong, not just the status code. Catches soft-error 200s.
<DNS-FAIL> <TLS-FAIL> <TO>Printed in exactly the column that broke, so you see where it failed, not just that it did.
autoVerify reason plus the offending certificate, read off the connection that failed. No re-run.
--tls-infoIssuer, SANs and days until expiry, colored as the deadline gets close.
-STimes every chunk as it arrives. CHUNKS, AVG_GAP and MAX_GAP show stalls mid-stream.
-S --statsDistribution over every inter-chunk gap. Meaningful at -c 1, and it spots proxy batching.
-P / -p IPHold every run on one backend instead of measuring a different one per DNS round-robin.
--server-hintsWhich server, edge, CDN or backend answered each request, flagged constant or varied.
--show-headers --capture-headerCache HIT/MISS verdict, plus any response header you name, tracked per request.
REDIRECTHop count and time in their own column, so TOTAL_TIME always adds up.
--http2The PROTO column shows what was actually negotiated: h1, h1.0, h2 or h3.
-b -j --show-cookiesCookies persist across every run in an invocation. Test login flows and session affinity.
--capture-on failedFull responses written to disk for the runs that matter, without moving the timings.
-H -d -X -4 -6Auth headers, POST bodies, methods, IP family. Your muscle memory already works.
--prometheusRe-probes on every scrape. Helm chart and ServiceMonitor included.
ghcr.ioProbe from a pod, CI job or bastion inside the network you're debugging. No pycurl build.
catppuccinCool blues for fast, warm peach and red for slow. Auto-off when piped.
A single Python script on top of pycurl. Or skip the build and use the image.
# pycurl must match libcurl's TLS backend
git clone https://github.com/bytebeast/check-endpoint.git
cd check-endpoint
python3 -m venv ~/.venvs/check-endpoint
source ~/.venvs/check-endpoint/bin/activate
PYCURL_SSL_LIBRARY=openssl pip install pycurl
./check-endpoint.py -c 10 --stats https://example.com# nothing to install beyond Docker
docker run --rm -it \
ghcr.io/bytebeast/check-endpoint:3 \
-c 10 --stats https://example.com# probe from inside the cluster
kubectl run check-endpoint --rm -it -q --restart=Never \
--image=ghcr.io/bytebeast/check-endpoint:3 \
--command -- sh -c 'sleep 2; exec check-endpoint "$@"' \
-- --tls-info https://my-svc.my-ns.svc.cluster.local/health