On Mon, Jan 12, 2009 at 12:03:31PM +0100, Rainer Jung wrote: > On 12.01.2009 11:19, Rainer Jung wrote: >> On 12.01.2009 10:04, Joe Orton wrote: >>> Sending SIGTERM to the rotatelogs process and having the parent recycle >>> it should have done that already, surely, without adding all this >>> complexity to rotatelogs? >> >> Thanks for the hint. In fact I just tested it and it does work for >> rotatelogs used in CustomLog, but not for the ErrorLog one. That one >> simply dies, no automatic restart and further error messages are lost >> (quickly tested on Solaris, and only with 2.2.x, needs some further >> testing for trunk). > > It's the same for trunk: CustomLog automatically gets respawned, > ErrorLog not. ...
Sorry, my mistake, I hadn't realised the piped error-logs code was different in this respect. Your analysis looks spot on here anyway. > Does it make sense to add the reliable piped logs way of handling > loggers to the error log? Or is there any known dependency between the > code used in ap_open_piped_log() and the error log itself? Well, I certainly had the expectation that piped error-loggers were automatically restarted; I don't know if there's some technical reason why it's not done, I expect it's just an oversight. Regards, Joe