 What's the performance impact of logging in #PHP? Is there an I/O difference between writing to `STDERR` or `STDOUT` vs. a local file on disk? I've heard folks like @rasmus speak in the past about ensuring you don't have errors because, even if you have error logging set to `error_reporting(0);` (i.e., "off"), the error handler is still invoked, so there's still a tiny a perf. hit, but I'm curious about the I/O impact between the various log streams. 14 comments
@heiglandreas @rasmus Well, yeah. Iβm trying to understand how much of an impact it makes, even if minor. Iβm also curious whether this is low-hanging fruit that could significantly impact performance of servers under high load.  @ramsey The "significantly impact performance of servers under high load" is the thing where I believe you got other problems. Mostly when you got high load you are in a distributet setup where the underlying stream might not necessarily be a "real" filesystem so the minimal performamce gain will still cause network trafic with all the slowness that is tied to it. But yeah. Investigating is cool and perhaps it can shave off a cPU cycle or two.... /cc @rasmus @heiglandreas I understand. I still want to know what the performance impact is. @ramsey As I said! Interesting question from an academic point of view! Go and figure it out! And perhaps you can indeed make it a bit faster. @heiglandreas That's why I'm asking for pointers to resources to help me figure it out. ;-) @ramsey Thanks! TIL that STDOUT and STDERR are by default different in regards to buffering.... π @heiglandreas @ramsey This is a very interesting question, especially considering the thousands of extra notices and warnings that 8.2 is generating on servers hourly. @e3betht @heiglandreas This comes closest to the type of resource I'm looking for. I might need to replicate their benchmarks for Java in PHP, using comparable libraries/approaches. https://blog.sebastian-daschner.com/entries/logging-performance-comparison  @ramsey @rasmus I worked a few years with logging on legacy systems. My approach was writing into rotating files (some were minutely with max 10 files because of the stream size). Filebeat streaming to logstash parsing to elastic (all in private network on VMs). The only problems I had to face was managing the free space. I used the rotating logs on FS as a broker. I donβt really know the impact of file logging as I did, but this was negligible in comparison with all the legacy bad designs. |
Gregory's Server
@ramsey From an academic point that seems an interesting question indeed!
From a performance point of view I'd say "when that is your problem, you got other problems"
/cc @rasmus