Repository navigation
Conversation
The amount of allocated system memory is kept in `real_size`, including the allocated `cached_chunks`. Thus, we need to keep the proper count at the end of the shutdown.
|
I think, it's right now. |
|
@cmb69 @dstogov I believe this change has had an (unforeseen?) side effect which can be relevant in some scenarios. Within a single php worker process, a previous request will now more directly impact the amount of memory available in the following request. I believe the new behaviour is more correct, as the memory reported now more accurately reports the actual consumed memory from the OS. Our code can also be adapted to handle this new behaviour, probably by taking the currently available memory into account. So there is probably nothing actionable (except maybe more general improvements to the memory manager), but this reply can maybe be useful as additional documentation. In our case we call The following script illustrates the issue. Works with apache's mod_php + mpm_prefork (making sure only 1 worker runs), but is more easily tested with php's internal webserver ( PHP 8.0.14, the memory reported goes down for each executed request (which is incorrect): PHP 8.0.15. Memory reset at start will fail when the memory of the previous request is not collected. |
|
I hope #8053 should fix the problem |
The amount of allocated system memory is kept in
real_size, includingthe allocated
cached_chunks. Thus, we need to keep the proper countat the end of the shutdown.