Keep the log fd, don't re-open logfile in every logs

[enjoy-binbin]

Currently in serverLog, we fopen and fclose the logfile in every call, if the log prints too much, it will waste performance. Also frequent open may cause blocking in some caes, like doing a big rename in the same dir (like rename the bgsave temp RDB).

In this PR, we keep the log fd, and only re-open it when dir changed. This also improve performance, using benchmark DEBUG log shows that it improves the performance by 3 times.

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 90.32258% with 3 lines in your changes missing coverage. Please review.

Project coverage is 70.35%. Comparing base (370bdb3) to head (3c37799). Report is 288 commits behind head on unstable.

Files with missing lines	Patch %	Lines
src/config.c	60.00%	2 Missing :warning:
src/server.c	95.83%	1 Missing :warning:

Additional details and impacted files

@@            Coverage Diff            @@
##           unstable     #906   +/-   ##
=========================================
  Coverage     70.34%   70.35%           
=========================================
  Files           112      112           
  Lines         61490    61492    +2     
=========================================
+ Hits          43258    43265    +7     
+ Misses        18232    18227    -5     

Files with missing lines	Coverage Δ
src/debug.c	54.37% <100.00%> (+0.39%)	:arrow_up:
src/server.h	100.00% <ø> (ø)
src/server.c	88.56% <95.83%> (+0.02%)	:arrow_up:
src/config.c	78.93% <60.00%> (+0.23%)	:arrow_up:

... and 9 files with indirect coverage changes

[enjoy-binbin]

@madolson @zuiderkwast @PingXie do you guys want to takes a look with this? i would like to include it in the RC3

[madolson]

This is a breaking change IIRC, if there are existing mechanisms for log rotation we need to send a signal to the process to close and re-open the log. We discussed this at length in Redis, but never committed to it.

[zuiderkwast]

Good point @madolson. I think we can't do this change without a more advanced solution then.

We can use ionotify on linux and kqueue on BSD/MacOS to watch the file and be notified when it is renamed. Then we close and reopen it. Fallback to the old behaviour if no watch method is available.

[enjoy-binbin]

ok, i see the point, indeed in this case, we need to find a way to re-open the file.

[PingXie]

This is a breaking change IIRC, if there are existing mechanisms for log rotation we need to send a signal to the process to close and re-open the log. We discussed this at length in Redis, but never committed to it.

That is great information. I think it would be very helpful if we could leave this context in serverLogRaw.