> cluster.data-self-heal-algorithm (default reset) > > Selects between "full", "diff", and “reset”. The "full" algorithm copies the > entire > file from source to sinks. The "diff" algorithm copies to sinks only those > blocks > whose checksums don't match with those of source. Reset uses a heuristic > model. If the file does not exist on one of the subvolumes, or a zero-byte > file > exists (created by entry self-heal) the entire content has to be copied > anyway, so > there is no benefit from using the "diff" algorithm. If the file size is > about the > same as page size, the entire file can be read and written with a few > operations, > which will be faster than "diff" which has to read checksums and then read and > write. > > > cluster.self-heal-window-size (default 16) > > Specifies the maximum number of blocks per file on which self-heal would > happen simultaneously. > > > > maybe cluster.data-self-heal-algorithm = diff could help ?
The default 'reset' seems reasonable. But it needs to compute checksums over the whole files, and that produces this high load. _______________________________________________ pve-devel mailing list [email protected] http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
