./bombardier --version
bombardier version v1.2.5 darwin/amd64
What operating system and processor architecture are you using (if relevant)?
darwin/amd64
What did you do?
I use bombardier to make some load tests of a remote service and verify that the service meets the performance expectations.
The service is exposed in http2 (can fallback to http1 thanks to alpn) and the bombardier load test is run as follow:
Everything is working well until I try to check the behavior of my service during a rolling update (check that there is no loss, errors , latency or throughput degradation).
During a rolling update, each pod is deleted 1 by 1 and active connection is gracefully shutdown by sending a goAway frame. All pending requests are honored. Everything is made to keep the service reachable during the rolling update.
The result of the bombardier command shows the following error:
Statistics Avg Stdev Max
Reqs/sec 375273.79 118131.95 565257.94
Latency 46.94us 429.41us 247.28ms
Latency Distribution
50% 10.00us
75% 28.00us
90% 109.00us
95% 199.00us
99% 486.00us
HTTP codes:
1xx - 0, 2xx - 2241, 3xx - 0, 4xx - 0, 5xx - 0
others - 11255262
Errors:
Get "https://my-remote-http2-service.com": http2: no cached connection was available - 11255262
What you expected to happen?
I was expecting bombardier to be able to create a new connection like it is able to do with http1.1 when it receives a Connection: close header (the same process with --http1 option works like a charm without any error).
What actually happened?
The goAway frame seems received and well interpreted but bombardier keeps trying to reuse the connection which seems to lead to the http2: no cached connection was available error.
What version of bombardier are you using?
What operating system and processor architecture are you using (if relevant)?
darwin/amd64What did you do?
I use bombardier to make some load tests of a remote service and verify that the service meets the performance expectations. The service is exposed in http2 (can fallback to http1 thanks to alpn) and the bombardier load test is run as follow:
./bombardier -c 20 -d 30s --http2 -k -l -m GET https://my-remote-http2-service.comEverything is working well until I try to check the behavior of my service during a rolling update (check that there is no loss, errors , latency or throughput degradation).
During a rolling update, each pod is deleted 1 by 1 and active connection is gracefully shutdown by sending a
goAwayframe. All pending requests are honored. Everything is made to keep the service reachable during the rolling update.The result of the bombardier command shows the following error:
What you expected to happen?
I was expecting bombardier to be able to create a new connection like it is able to do with http1.1 when it receives a
Connection: closeheader (the same process with --http1 option works like a charm without any error).What actually happened?
The
goAwayframe seems received and well interpreted but bombardier keeps trying to reuse the connection which seems to lead to thehttp2: no cached connection was availableerror.Thanks for you help.