UNIX scheduler product

Tue Jun 15 05:26:55 PDT 2004

On Mon, 14 Jun 2004, richard childers / kg6hac wrote:

> What's wrong with a series of crontab entries, interoperation 
> coordinated through use of lock files, made visible via NFS? All 
> rock-solid technology, decades old now.

*shudder*

"lock files" over nfs are notoriously difficult if not impossible
to do correctly due to a variety of reasons, most notably nfs client
cacheing complications and implementation bugs/quirks. also, nfs does
not completely support the the same semantics as a local filesystem
(e.g. O_EXCL) and such support varies with nfsv2 vs. nfsv3, so you must
be aware of that when choosing the lockfile method. if you google around
there are some "dot-locking" algorithms which are purportedly nfs-safe
(there is such code in exim which claims to be largely nfs-safe), but
who knows, i've seen many a buggy nfs implementation.

also, lock files are also prone to becoming stale when the lockers crash
or are otherwise buggy, further complicating this mechanism.

cron could really use a serialization feature, e.g. don't invoke the next
job until the previous one has completed.  i've seen a number of scripts
meant to be invoked by cron include this mechanism, but it seems this
is required often enough that it makes sense for cron to handle this
rather than the individual job.