We did a software change which increased Apcu usage on a production server running perfectly for months.
After about 20 hours of running with response time of 15msec at ~400 hits/sec the server response time increases to ~500msec. (This repeats itself after ~20 hours if we restart apache or reset apcu)
Resetting apcu with apc_clear_cache ("user"); brings the response time back to normal so the problem is APCu overload of some kind.
How can this be fixed? There is nothing in the logs explainng what the bottleneck is
Details:
At the time the issue happens the data on apcu usage is:
APCu memory consumption: 256,851,984 bytes
APCu items number: 519,311
Server:
PHP Version 7.3.18-1+ubuntu18.04.1+deb.sury.org+1
Apache/2.4.29 (Ubuntu)
APCu Support Enabled
Version 5.1.17
APCu Debugging Disabled
MMAP Support Enabled
MMAP File Mask no value
Serialization Support php
Build Date Aug 27 2019 08:46:38
Directive Local Value Master Value
apc.coredump_unmap Off Off
apc.enable_cli Off Off
apc.enabled On On
apc.entries_hint 4096 4096
apc.gc_ttl 3600 3600
apc.mmap_file_mask no value no value
apc.preload_path no value no value
apc.serializer php php
apc.shm_segments 1 1
apc.shm_size 1024M 1024M
apc.slam_defense Off Off
apc.smart 0 0
apc.ttl 0 0
apc.use_request_time On On