feat: support retry in health-check #131
Conversation
Signed-off-by: yiyiyimu <wosoyoung@gmail.com>
Yiyiyimu
force-pushed
the
feat/health-check-retry
branch
from
a6c14c2 to
3daee06
Compare
Yiyiyimu
changed the title
Signed-off-by: yiyiyimu <wosoyoung@gmail.com>
| end | ||
|
|
||
| self.init_count = (self.init_count or 0) + 1 | ||
| local pos = self.init_count % endpoints_len + 1 |
There was a problem hiding this comment.
Is it intended to skip the first endpoint? As the minimal init_count here is 1.
There was a problem hiding this comment.
I also failed to figure out the purpose of this. It's the implementation of the earliest v3 API.
Hi @nic-chen do you remember why is it set so
There was a problem hiding this comment.
how about self.init_count = (self.init_count or -1) + 1 ?
There was a problem hiding this comment.
I also don’t remember, it’s contributed by the community
There was a problem hiding this comment.
Got it. Fixed as suggested. Thanks!
lib/resty/etcd/v3.lua
Outdated
| elseif body.error and body.error.http_code >= 500 then | ||
| if health_check.conf ~= nil then | ||
| -- health_check retry should do nothing here | ||
| -- and let connection broke to create a new one |
There was a problem hiding this comment.
Actually I mean closed 🤣 fixed
| return _request_uri(self, endpoint, "GET", | ||
| endpoint.http_host .. "/v2/stats/self", | ||
| nil, self.timeout) | ||
| return _request_uri(self, "GET", "/v2/stats/self", nil, self.timeout) |
There was a problem hiding this comment.
Not about this PR: is there /v2 API in the /v3 file?
There was a problem hiding this comment.
not for API exists in both version, like /get and /put
Signed-off-by: yiyiyimu <wosoyoung@gmail.com>
Signed-off-by: yiyiyimu <wosoyoung@gmail.com>
…oint choose problem in etcd (https://github.com/api7/lua-resty-etcd/pull/131\#discussion_r655804238) Signed-off-by: yiyiyimu <wosoyoung@gmail.com>
Signed-off-by: yiyiyimu wosoyoung@gmail.com
retryin health-check. When it is enabled, operations would keep trying the same endpoint till unhealthy, and then other endpoints, until success or no endpoints healthy. So we could ensure each operation would send to etcd with best effort.Currently not support for
read_watchsince it would be better to close the original connection and create a new watch with a new endpoint.choose_enpointintorequest_uriandrequest_chunkso it could find alternative endpoint easier