fix envoy config to avoid overflows #309
Conversation
Signed-off-by: Alexander Lukyanchenko <digwnews@gmail.com>
Signed-off-by: Alexander Lukyanchenko <digwnews@gmail.com>
Signed-off-by: Alexander Lukyanchenko <digwnews@gmail.com>
|
Thank you. I'm worried about setting these limits to arbitrary values but I think for 0.4.1 we can land this as is and take a p2 issue to bikeshed their correct values. |
| max_connections: 100000 | ||
| max_pending_requests: 100000 | ||
| max_requests: 60000000 | ||
| max_retries: 50 |
There was a problem hiding this comment.
I'm not sure about this value. What I see in testing is
contour::127.0.0.1:8001::cx_active::1
contour::127.0.0.1:8001::cx_connect_fail::2
contour::127.0.0.1:8001::cx_total::3
contour::127.0.0.1:8001::rq_active::514
contour::127.0.0.1:8001::rq_error::4
contour::127.0.0.1:8001::rq_success::0
contour::127.0.0.1:8001::rq_timeout::0
contour::127.0.0.1:8001::rq_total::518
contour::127.0.0.1:8001::health_flags::healthy
contour::127.0.0.1:8001::weight::1
contour::127.0.0.1:8001::region::
contour::127.0.0.1:8001::zone::
contour::127.0.0.1:8001::sub_zone::
contour::127.0.0.1:8001::canary::false
contour::127.0.0.1:8001::success_rate::-1
518 in progress connections, but nothing has timed out yet. Retrying is fine, even retrying indefinitely is fine, but I think a reasonably short connection timeout, say < 5 seconds, should be applied to each attempt.
It doesn't matter so much for this change as we're going to embiggen the max incoming connections on the server side in #308, but we should revisit it in #313
|
Thank you very much for working on this. I've tested this plys #308 and confirmed its raised the limits as we expected. I've squashed this because I need to backport it to the 0.4 branch. |
* fix envoy config to avoid overflows Signed-off-by: Alexander Lukyanchenko <digwnews@gmail.com>
This patch uses rest mapper instead of `client.Get()` to verify Gateway CRDs. It reduce API calls and is same method with Contour. Fix projectcontour/contour-operator#270 Signed-off-by: Kenjiro Nakayama <nakayamakenjiro@gmail.com>
Fixes #291
Not sure that we need to expose these parameters to be configurable. If you think so, let me know, I'll expose them.
It's necessary for big clusters.