Yichun Zhang
Yichun Zhang
@bakins Creating new closures at every request does introduce some overhead (observable in on-CPU time Flame Graphs, for example), so I think we need the ability to pass user arguments...
@bakins Creating closures per request could be a real bottleneck, as observed as 20% ~ 30% overall overhead in an on-CPU time flame graph for our Lua WAF system. Even...
What do you mean by "connection used time"? Let's talk with code snippets to avoid any confusions from vague natural language descriptions.
Hello! On Sat, Feb 2, 2013 at 6:56 PM, Jose Diaz-Gonzalez wrote: > > Useful in contexts where the cosocket api is not available - such as > init_by_lua. >...
@josegonzalez I'd like to add a simple blocking implementation of the cosocket API for `init_by_lua*` (it does not matter at all to block there). Are you willing to contribute a...
@josegonzalez No, not really. The current implementation is nonblocking.
Our current plan is to directly run an event loop in the `init_by_lua*` context and make cosockets directly work there.
@bumi001 Mind to contribute a pull request for this? Thanks!
@eulhaque It is safe to use in production. The "experimental" word just means the API might get changed in the future.
The random failures can be reproduced by setting the env `TEST_NGINX_SLEEP=0.001`. The tests should be made more robust.