I'm running HHVM 3.1.0 via nginx 1.6.0 on a CentOS 6.4 system.
I know when you load a php file for the first time, (or first few time depending on the number of JIT warmup passes you configure to wait) the response is slow, but once compilation has occurred, subsequent requests are much faster.
One thing I noticed is once the request have speed up, they stay fast for a while, but if I wait overnight and try to run requests again, it performs as if the code was never compiled--just like restarting the HHVM daemon.
Does HHVM have something like a 'compile lifetime' that re-compiles code after a certain number of requests or time? If so, is there any setting way to control this? Obviously, on production machines once your cache is warmed up and all you PHP code has been JIT compiled, you don't want to incur these cost again until code changes.
For reference, I've included my HHVM config ini file:
; php options
pid = /var/run/hhvm/pid
date.timezone="America/New_York"
; hhvm specific
hhvm.server.file_socket=/var/run/hhvm/hhvm.sock
;hhvm.server.port = 9000
hhvm.server.type = fastcgi
hhvm.server.default_document = index.php
hhvm.log.level = Warning
hhvm.log.always_log_unhandled_exceptions = true
hhvm.log.runtime_error_reporting_level = 8191
hhvm.log.use_log_file = true
hhvm.log.file = /var/log/hhvm/error.log
hhvm.log.header = true
hhvm.repo.central.path = /var/run/hhvm/hhvm.hhbc
hhvm.mysql.typed_results = false
hhvm.eval.jit_warmup_requests = 0
Ok, giant egg on my face. The issue I was having was with the cache in the database, not HHVM. On restart (like during backup) the Mysql query cache gets lost, hence overall page slow down. Sometimes it's right in front your face...