On Mon, Sep 12, 2016 at 11:00 AM, Ilya Moldovan <il.moldovan@xxxxxxxxx> wrote: > Thanks, John > > But why listing files in a directory with about a million files takes > about 30 minutes? Unless you've enabled experimental features, you're trying to read in a single 1-million-inode directory object — and it sounds like the MDS is succeeding. But by default the MDS cache size is only 100k dentries, so it immediately trims most of the directory out of cache; then the client submits a request for the next chunk and it does it again, etc etc. So, assuming you have the RAM, try setting "mds cache size = 2000000" or something, then see how it behaves. And consider not trying to create directories that large if you don't need to — we have plans for handling it much more gracefully in the future (in fact the code is one of the experimental features) but it's not enabled in Jewel. -Greg > > Ilya Moldovan > > 2016-09-08 20:59 GMT+03:00, Ilya Moldovan <il.moldovan@xxxxxxxxx>: >> Hello! >> >> How CephFS calculates the directory size? As I know there is two >> implementations: >> >> 1. Recursive directory traversal like in EXT4 and NTFS >> 2. Calculation of the directory size by the file system driver and save it >> as an attribute. In this case, the driver catches adding, deleting and >> editing files on the fly and changes the size of the directory. In this >> case there is no need recursive directory traversal. >> >> The directory which we are requesting a size can potentially contain >> thousands of files at different levels of nesting. >> >> Our components will call the the directory size using the POSIX API. The >> number of calls of this attribute is will be high and recursive directory >> traversal is not suitable for us. >> >> Thanks for the answers! >> > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |