On Tue, Mar 11, 2014 at 1:23 PM, Ivan Zhakov <i...@visualsvn.com> wrote:
> On 8 March 2014 01:35, <stef...@apache.org> wrote: > > Author: stefan2 > > Date: Fri Mar 7 21:35:54 2014 > > New Revision: 1575418 > > > > URL: http://svn.apache.org/r1575418 > > Log: > > Enable FSFS format 7 repositories to be packed without completely > > blocking commits. We simply introduce a separate lock file for > > 'svnadmin pack' and take out the global write lock for packing > > revprops and switching the shard to "packed" state. > > > > Most of the run time is spent building the revision pack file > > and does not require any synchronization with writers. > > > [...] > > > @@ -1987,10 +2005,33 @@ svn_fs_fs__pack(svn_fs_t *fs, > > apr_pool_t *pool) > > { > > struct pack_baton pb = { 0 }; > > + fs_fs_data_t *ffd = fs->fsap_data; > > + svn_error_t *err; > > + > > pb.fs = fs; > > pb.notify_func = notify_func; > > pb.notify_baton = notify_baton; > > pb.cancel_func = cancel_func; > > pb.cancel_baton = cancel_baton; > > - return svn_fs_fs__with_write_lock(fs, pack_body, &pb, pool); > > + > > + if (ffd->format >= SVN_FS_FS__MIN_PACK_LOCK_FORMAT) > > + { > > + /* Newer repositories provide a pack operation specific lock. > > + Acquire it to prevent concurrent packs. */ > > + apr_pool_t *subpool = svn_pool_create(pool); > What is the reason for using subpool here? Could you please document > it if any, otherwise it looks very confusing. > Done in r1576427. > > + const char *lock_path = svn_dirent_join(fs->path, > PATH_PACK_LOCK_FILE, > > + subpool); > > + err = svn_fs_fs__get_lock_on_filesystem(lock_path, subpool); > > + if (!err) > > + err = pack_body(&pb, subpool); > > + > > + svn_pool_destroy(subpool); > > + } > > + else > > + { > > + /* Use the global write lock for older repos. */ > > + err = svn_fs_fs__with_write_lock(fs, pack_body, &pb, pool); > > + } > > + > > + return svn_error_trace(err); > > } > Thanks for the feedback! -- Stefan^2.
