It seems to me that there are at least three improvements which
should be made to copyfile:
1. If the file on the overlord and minion are the same, that file should not
be copied. This saves bandwidth/processing time.
2. Overwriting the file on the minion should be atomic, e.g. the file on the
overlord should be copied to a tempfile on the minion then moved to overwrite
the target file as a final step. Right now if funcd dies or something goes
wrong mid-copy you end up with half a file on the minion.
3. There should be an option (off by default) such that if the file which has
been copied over to the minion is not identical to the file on the overlord it
is re-copied, or at least does not overwrite the file on the minion. This
won't work for some files (busy logfiles) but for most files it seems like an
extra bit of reliability which most people would want in most situations.
I submitted a patch (though not a git-patch... my bad!) back in
January 2011 which implements #1, but imperfectly. The overlord still has to
send a (small) packet to the minion for each 60KB chunk of target file.
Would anyone be bothered terribly if I worked on #2 and #3? While I am
at it I will try to find a way to fix up my patch for #1 as well. There is
some overlap between these patches. Both will involve calculating an MD5 hash
of the file on the overlord, for example.
P.S. Is the new mailing list on fedorahosted.org live yet? I am
sending this message to both just in case.
--
Marcus
_______________________________________________
func mailing list
func@xxxxxxxxxxxxxxxxxxxxxx
https://lists.fedorahosted.org/mailman/listinfo/func
