mirror of https://github.com/zulip/zulip.git
nginx: Make uwsgi timeout shorter than nginx-to-uwsgi timeout.
The nginx-to-uwsig-timeout defaults to 60s, which is exactly the same as the current "harakiri" timeout configured in uwsgi (which limits the length a request can run before the worker is terminated). This causes a race, where if nginx hits its 60s before uwsgi, then we return a 504; otherwise, we get a 502. Make the nginx-to-uwsgi timeout explicit, and shorten the "harakiri" timeout to be explicitly less than that. Document the 60s timeout, which all outer reverse proxies must be set to _longer than_ in order to have proper "onion" timeouts.
This commit is contained in:
parent
3209086370
commit
1bd0ab506c
|
@ -255,13 +255,13 @@ things you need to be careful about when configuring it:
|
||||||
browsers. This [nginx code snippet][nginx-proxy-longpolling-config]
|
browsers. This [nginx code snippet][nginx-proxy-longpolling-config]
|
||||||
does this.
|
does this.
|
||||||
|
|
||||||
The key configuration options are, for the `/json/events` and
|
The key configuration options are:
|
||||||
`/api/1/events` endpoints:
|
|
||||||
|
|
||||||
- `proxy_read_timeout 1200;`. It's critical that this be
|
- `proxy_read_timeout 1200;`. It's critical that this be significantly above
|
||||||
significantly above 60s, but the precise value isn't important.
|
60s, but the precise value isn't important. This is most important for the
|
||||||
- `proxy_buffering off`. If you don't do this, your `nginx` proxy may
|
events API, but must be applied to all endpoints.
|
||||||
return occasional 502 errors to clients using Zulip's events API.
|
- `proxy_buffering off`. If you don't do this, your `nginx` proxy may return
|
||||||
|
occasional 502 errors to clients using Zulip's events API.
|
||||||
|
|
||||||
1. The other tricky failure mode we've seen with `nginx` reverse
|
1. The other tricky failure mode we've seen with `nginx` reverse
|
||||||
proxies is that they can load-balance between the IPv4 and IPv6
|
proxies is that they can load-balance between the IPv4 and IPv6
|
||||||
|
|
|
@ -18,4 +18,8 @@ uwsgi_param HTTP_X_FORWARDED_PROTO $trusted_x_forwarded_proto;
|
||||||
uwsgi_param HTTP_X_FORWARDED_SSL "";
|
uwsgi_param HTTP_X_FORWARDED_SSL "";
|
||||||
uwsgi_param HTTP_X_PROXY_MISCONFIGURATION $x_proxy_misconfiguration;
|
uwsgi_param HTTP_X_PROXY_MISCONFIGURATION $x_proxy_misconfiguration;
|
||||||
|
|
||||||
|
# This value is the default, and is provided for explicitness; it must
|
||||||
|
# be longer than the configured 55s "harakiri" timeout in uwsgi
|
||||||
|
uwsgi_read_timeout 60s;
|
||||||
|
|
||||||
uwsgi_pass django;
|
uwsgi_pass django;
|
||||||
|
|
|
@ -47,7 +47,7 @@ auto-procname=true
|
||||||
procname-prefix-spaced=zulip-django
|
procname-prefix-spaced=zulip-django
|
||||||
|
|
||||||
# Longest response allowed, in seconds, before killing the worker
|
# Longest response allowed, in seconds, before killing the worker
|
||||||
harakiri=60
|
harakiri=55
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
Loading…
Reference in New Issue