From: Peter Dimov (pdimov_at_[hidden])
Date: 2021-03-04 19:56:14
Andrey Semashev wrote:
> On 3/4/21 10:04 PM, Peter Dimov via Boost wrote:
> > Andrey Semashev wrote:
> >
> >> Unfortunately, text_multifile_backend is supposed to open and close
> >> file on every log record, as the file name is generated from the log record.
> >
> > If you maintain a small LRU cache, will it help?
>
> That's what I had in mind, but it will have its own problems.
>
> One of the main use cases for text_multifile_backend is when you want to
> maintain a separate log for each of the user's business process entity (e.g. a
> session or a connection of some sort). There can be any number of such
> entities, and I can't come up with a reasonable limit for a cache in the sink
> backend. Naturally, the cache will be ineffective if the number of actively
> logging entities exceeds its capacity. At the same time I don't want to keep any
> stale open files in the cache.
>
> It seems, I'll have to maintain some sort of a timeout for closing files and
> cleaning up the cache. But I will only be able to perform the cleanup on log
> records, which means I'll still have stale open files in cache if there's no logging
> happening.
>
> Keeping open files is not only a concern from resource usage perspective. It
> may affect user's experience, as e.g. on Windows he won't be able to remove
> the open files or the containing directory.
> Deleting an open file on POSIX systems will succeed, but it will also mean some
> subsequent log records can be lost as they will be written to the deleted file
> instead of the new one.
It's definitely not trivial.
The right thing to do would probably be to implement a deferred close where a
file lingers around for a second or so, then is closed by a background thread. If
it's used again within this second, it doesn't need to be closed and reopened.