2011-03-20 Andreas J. Koenig * More options to pass through rrr-client * More tests * .recent/ or .rrr/ for metadata * memory leak in rrr-server * lockdirectory expiration? server died and blocked fsck for so long. * rewrite _thaw_without_pathdb as a forked child job to have it format-independent * profiling! Rewrite slow parts in C. * rrr-server probably not robust under all possible conditions; maybe add some regular fsck or something. Consider the case of IN_Q_OVERFLOW again. * signal handlers * how would an rsync-free HTTP variant look like? See 2008-10-10 again. * are we sure we do NOT LEAVE DOT FILES around? * ---- no open todos below this line ---- 2011-02-21 Andreas J. Koenig * rrr-server has a memory leak root 16063 94.7 10.6 358404 354600 pts/21 RN Feb18 4117:16 /home/src/perl/repoperls/installed-perls/perl/v5.13.8-16-gf53580f/bin/perl -Ilib bin/rrr-server --verbose /home/ftp/incoming/RECENT-1h.yaml experimenting with a fork for aggregate(): root 12196 60.2 0.2 13652 9920 pts/21 RN 08:25 1:56 /home/src/perl/repoperls/installed-perls/perl/v5.13.8-16-gf53580f/bin/perl -Ilib bin/rrr-server --verbose /home/ftp/incoming/RECENT-1h.yaml A delete of several hundred files: root 12196 72.8 0.3 13792 10056 pts/21 RN 08:25 3:48 /home/src/perl/repoperls/installed-perls/perl/v5.13.8-16-gf53580f/bin/perl -Ilib bin/rrr-server --verbose /home/ftp/incoming/RECENT-1h.yaml Delete several hundred again: root 12196 79.8 0.3 14788 10944 pts/21 RN 08:25 6:34 /home/src/perl/repoperls/installed-perls/perl/v5.13.8-16-gf53580f/bin/perl -Ilib bin/rrr-server --verbose /home/ftp/incoming/RECENT-1h.yaml It's a lot. 2011-02-20 Andreas J. Koenig * start consistently considering the RECENT files themselves not part of the set. 2011-02-19 Andreas J. Koenig * Found the term anonymous one way file system * No support for empty directories 2011-02-17 Andreas J. Koenig * ALERT rrr: we have a piece of code somewhere in the mirror() subroutine that reads the raw YAML file with normal open() and cuts it off for efficiency, then feeds it to yaml::load in order to get some metadata. Reading the thing with the whole tail of the recent array kills the whole idea. The whole thing will not work with JSON. So we must invent a protocol 2 that works with one additional index files, say "A", only containing meta. 2011-02-09 Andreas J. Koenig * I think interval guaranteed with the $last_aggregate_call variable (both in fsck and server) need to be fixed with a $recc variable to something like 60 seconds or even less. * minor bug: fsck just added the lockfile to the index which should be considered bookkeeping. * speed: a large directory remove that is done by the kernel in 1 seconds triggers minutes of bookkeeping (07:56:42 - 08:01:10 for ~3000 files, all due to Schlemiehl again. Want to do some collecting of pending ops before actually writing to the RECENT files. Lock! Repeating the timing with 2600 files and it took 21:28:08 - 21:35:27. Now having rewritten the loop to use batch_update(): 4383 file removed 21:43:37 - 21:44:12. From 7:30 to 0:35 while doing 60% more work that's a joy. * At the moment rrr-init is not needed anymore, rrr-server can be started and followed by rrr-fsck. Still stupid but not the main showstopper. 2011-02-05 Andreas J. Koenig * IN_Q_OVERFLOW needs to trigger an fsck, likewise the entry into the server. 2010-10-26 Andreas J. Koenig * forking slave is done, needs more testing. 2010-07-06 Andreas J. Koenig * The two next most important goals are to get a useable rrr-server and a forking slave implementation that uses virtually no memory after the first pass through. Does it need the status file? * rrr-init now being fast we can now remove the alpha alerts. Done. 2010-06-16 Andreas J. Koenig * rrr-init seems to be sssslllloooowwww. At 3:49 I started it for the video tree. After twelve minutes the file has reached a size of 500k. It is locking and rewriting the YAML after every file. Wasn't there an unlocked variant? For "init" one would not expect locked correctness and a rewrite after every item. Given that this tends to become slower with every file it is really unacceptable. Reminds me of rrr-dirtyupdate (the former name of rrr-update). It seems the code was not dirty enough and that was the reason why it was so slow. And rrr-update does not have a --dirty switch, so it looks like we have no way to force dirty and fast operation yet but we need it for init. Irritating the bootstrapping race: the order {(1) fill initial array, (2) start server} leaves the need for an fsck to cover the time between 1 and 2. Given that regular fsck is needed in any case the price is not high but still doesn't give a warm feeling. Maybe starting the server will by itself do an fsck? 2010-06-15 Andreas J. Koenig * remove the wild alpha alerts and say, it is alpha but parts of it are already working very well or so. 2009-12-10 Andreas J. Koenig * Prominent request for statusfile to safe memory consumption. I'd like to work backwards: get a verification program that verifies how well in sync two nodes are without using frmr itself. Pretend that the forking agent already works and can be interrupted with ^C. Then let it die at random and restart at random and see the memory consumption and the failure rate over time. Compare it with the curve we have now. Of course we also invented the status file as a sharp debugging tool. I'd like to see it in action. See also: F:R:M:Recent::thaw(). 2009-06-22 Andreas J. Koenig * Todo: rrr-init to generate some index files so that rrr-server can be started. 2009-05-02 Andreas J. Koenig * DONE: rename rrr-dirtyupdate to rrr-update and add an option --dirty so we can use it for normal update with a single file. And add a cronjob that runs rrr-fsck so we get a reminder when something is not working as it should. 2009-04-27 Andreas J. Koenig * Todo: I still see more link_stat errors than expected. I come back to the former suspicion that we do not do necessary bookkeeping on deleted files so that when we reach the "new" event on a meanwhile deleted file we try to mirror it although we could know better. It's just noise, no harm but noise is irritating. * Todo: more experimenting with runstatusfile to get it official * Todo: make sure $ENV{LANG}="C" when we call rsync because we will parse the error. rsync: link_stat "/authors/id/T/TH/THINC/DateTime-Format-Flexible-0.02.tar.gz" (in PAUSE) failed: No such file or directory rsync: link_stat "/authors/id/A/AG/AGENT/SSH-Batch-0.001.meta" (in PAUSE) failed: No such file or directory (2) And that ignore_link_stat_error is propagated in sparse_clone. DONE For the record: the bug was that ignore_link_stat_error was lost during sparse_clone. And then the second bug was that it did not default to true. Both are now fixed. 2009-04-25 Andreas J. Koenig * Todo: fill rrr program with life 2009-04-25 Andreas J. Koenig * Minor bug: I see complaints about files not existing like /authors/id/G/GW/GWADEJ/SVG-Sparkline-0.2.0.meta or /authors/id/R/RC/RCAPUTO/POE-1.004.readme In the RECENT files these are delete events but of course also new events in older recent files. I think we do not properly remember deletes when we skip-deletes and so do not filter them out when we later see the "new" event. Yes, but that what ignore_link_stat_error is for, we know about this race condition. We must make ignore_link_stat_error default to true. DONE * bug in Done.pm failing to merge two fields into one when a third is present. FIXED and accompanied with new test. 2009-04-24 Andreas J. Koenig * Todo: keep the index clean and run some kind of nondestructive fsck 4 times a day. * who is our backbone? http://cpan.cpantesters.org/authors/02STAMP # barbie http://cpan.solfo.com/authors/02STAMP # abh (http|ftp)://theoryx5.uwinnipeg.ca/pub/CPAN/ # rkobes 2009-04-17 Andreas J. Koenig * the changelog helper I included in the Makefile release memo does not show tags and is much less useful than I thought. gitk is probably much more convenient. 2009-04-16 Andreas J. Koenig * ABH writes: #### Sync 1239828608 (1/1/Z) temp .../authors/.FRMRecent-RECENT-Z.yaml- #### Ydr_.yaml ... DONE #### Cannot stat '/mirrors/CPAN/authors/.FRMRecent-RECENT-Z.yaml- #### Ydr_.yaml': No such file or directory at /usr/lib/perl5/site_perl/ #### 5.8.8/File/Rsync/Mirror/Recentfile.pm line 1558. #### unlink0: /mirrors/CPAN/authors/.FRMRecent-RECENT-Z.yaml-Ydr_.yaml is #### gone already at cpan-pause.pl line 0 Running without skip-deletes now but cannot reproduce. 2009-04-15 Andreas J. Koenig * consider state directory such we can restart after a ^C or come again from a cronjob. Reuse _runstatusfile: write atomically, see how we can use the dump to start running again. Do not leak and remove globs before writing. Do not only dump one rf, dump the whole complex. * consider the effect of ^C. Is it safe? * want to use Perl::Version to up the versions on every release. Only if things have changed, of course. Does this actually work? /usr/local/perl-5.10-g/bin/perl-reversion -dryrun /usr/local/perl-5.10-g/bin/perl-reversion -dryrun -bump lib/File/Rsync/Mirror/Recentfile.pm Must do the bookkeeping myself when this should happen, maybe based on git? I think I prefer the same version number for all pm files. But one also needs a routine for setting them. Ahh: /usr/local/perl-5.10-g/bin/perl-reversion -current 0.0.4 -set 0.0.5 lib/**/*.pm NEEDSMOREWORK * why indexer wrong? cpan[2]> m /Mirror::Recent/ Module = File::Rsync::Mirror::Recent (ANDK/File-Rsync-Mirror-Recent-0.0.4.tar.bz2) Module = File::Rsync::Mirror::Recentfile (ANDK/File-Rsync-Mirror-Recent-0.0.2.tar.bz2) Module = File::Rsync::Mirror::Recentfile::Done (ANDK/File-Rsync-Mirror-Recent-0.0.2.tar.bz2) Module = File::Rsync::Mirror::Recentfile::FakeBigFloat (ANDK/File-Rsync-Mirror-Recent-0.0.2.tar.bz2) 4 items found The META.yml was generated by ExtUtils::MakeMaker version 6.5101 The 0.0.2 release has a META.yml by ExtUtils::MakeMaker version 6.42 and without "provides". I just tried MM 6.5102 and it again produces only a provides for Recent, not for the other modules. My own fault. FIXED in the Makefile.PL. 2009-04-13 Andreas J. Koenig * Study http://en.wikipedia.org/wiki/Magnet_URI_scheme * is it true (as stated at https://fedorahosted.org/InstantMirror/wiki/ExistingRepositoryReplicationMethods) that rsync can lead to errors when upstream changes before client sync has completed? Can we see the error when we run this on upstream: % perl -e ' use Time::HiRes qw(time); while (){ open my $fh, ">", "changingfile.txt.new" or die; print $fh (time."\n") x 1000000; close $fh; rename "changingfile.txt.new", "changingfile.txt" or die; } ' And on the receiving end: % while true; do rsync k75:`pwd`/changingfile.txt .; cat changingfile.txt| uniq|wc -l sleep 1; done I cannot get it to fail. Apparently it is sufficient when upstream always writes atomically (which of course is mandatory)? SENT to mailing list Apr 28. * from ABH: > https://fedorahosted.org/InstantMirror/ > https://www.redhat.com/mailman/listinfo/instantmirror-list > irc.freenode.net channel #instantmirror A cool name for a project. Inspires me to write a few sentences about bittorrent's role in the grand picture. http://spreadsheets.google.com/pub?key=pGlWX10blP4u2kM05SDtiMg is a spreadsheet collected by Atul Aggarwal about bittorrent implementations. 2009-04-12 Andreas J. Koenig * Interesting last minute bug during real download testing: the output isn't as pretty anymore, it seems that more work is being done than needed. Introducing a highlevel _alluptodate method helped a lot in debugging this. FIXED now. Reason was that dirtymark on PAUSE currently is broken; it's different on 1h file than on the other files. This lead to frequent calls to _fullseed. Limiting the dirtymark check to $i==0 resolved the issue. 2009-04-12 Andreas J. Koenig * Todo: try if we can be faster with a native float package. I'm really glad to have a machine exposing perl floating point bugs, it forced me to retract this item from the changes file: "several speedups in the fakebigfloat code" before I get cpan testers fails. 2009-04-11 Andreas J. Koenig * Bug: when the dirtymark changes upstream, then the downstream server notices it immediately and mirrors the "1h" file but then leaves the rmirror loop and continues with "6h" on the next round through. This behaviour goes on until it reaches the Z file. Because we eat the Z file piecemeal we leave the loop after a while but after that everybody seems to have forgotten that there is some work left to be done. Bad, bad, bad. ETOOCONFUSING better name? Recentfile::get_remote_recentfile_as_tempfile OK Recentfile::resolve_recentfilename split_rfilename DONE Recent::_principal_recentfile_object _principal_recentfile_fromremote Recent::principal_recentfile OK Recent::_resolve_rfilename _principal_recentfile_fromremote_resosymlink some void, some not, some not void but called in void context. 2009-04-10 Andreas J. Koenig * Bug? When the dirtymark changes then the second tier hosts quickly reset their DONE state and restart mirroring. But then they mirror potentially outdated index files with inconsistent dirtymark because the first tier box may be a bit behind on the large recentfiles. This means we need a brake when iterating over the recentfiles that refuses to use a recentfile with the wrong dirtymark. 2009-04-08 Andreas J. Koenig * install the bin/* files and talk about rrr-overview somewhere. * Ask B. Hansen asks for tempdir accessor such that tempfiles get created outside the target tree. * Barbie asks for better logging capabilities 2009-03-29 Andreas J. Koenig * known bug: cannot configure the Recentfile objects to keep delete events. How to call the accessor for that? keep_delete_objects_forever? Sounds acceptable. FIXED (but untested). * known bug: keeps temporary index files lying around. 2009-03-24 Andreas J. Koenig * Some equivalent for for ( @{$rrr->recentfiles} ) { $_->verbose(0) } ? DONE 2009-03-22 Andreas J. Koenig * k81 is again client of k75 with different parameters. * broken now: (1) seeding and unseeding: rmirror is seeding and talking about it all the time and nobody reacts accordingly; (2) lots of temp files get created and not removed; culprit the new call to get_remote_recentfile_as_tempfile within Recent.pm; the manpage says the caller has to remove the tempfile after use. Need the drawing board. PARTLY FIXED: retracting the idea to call get_remote_recentfile_as_tempfile from Recent.pm and moving around seed and unseed such that uptodate now means "we have mirrored this recentfile's bunch of files" and seeded means "somebody believes the index file for this rf needs be refreshed". This seems to work now. Still collecting temp files that nobody cares to depollute. 2009-03-21 Andreas J. Koenig * Bug?: should it be harder than it is atm to set the timestamp to the future? YES, FIXED * bug with native integers: - epoch: 997872011 path: id/D/DE/DELTA/Crypt-Rijndael_PP-0.03.readme type: new - epoch: 1195248431 path: id/L/LG/LGODDARD/Tk-Wizard-2.124.readme type: new Native integer broke when native math was turned off. FIXED * Bug: something between id/P/PH/PHISH/CGI-XMLApplication-1.1.2.readme and id/C/CH/CHOGAN/HTML-WWWTheme-1.06.readme. Mirroring the Z file loops in this area. Yes, records out of order: 447003 - 447004 epoch: 995885533 447005 path: id/P/PH/PHISH/CGI-XMLApplication_0.9.3.readme 447006 type: new 447007 - 447008 epoch: 995890358 447009 path: id/H/HD/HDIAS/Mail-Cclient-1.3.readme 447010 type: new 447011 - 447012 epoch: 995892221 447013 path: id/H/HD/HDIAS/Mail-Cclient-1.3.tar.gz 447014 type: new FIXED with sanity check and later with the integer fix. * Bug: want the index files in a .recent directory * Bug: lots of dot files are not deleted in time * possible test case: can a delete change the timestamp? This would probably break the order of events. 2009-03-20 Andreas J. Koenig * 1233701831.34486 what's so special about this number/string? It belongs to id/G/GR/GRODITI/MooseX-Emulate-Class-Accessor-Fast-0.00800.tar.gz and atm lives in Y,Q, and Z. It is the first entry after id/--skip-locking which has timestamp 1234164228.11325 which represents a file that doesn't exist anymore. * another bug: Sync 1237531537 (31547/33111/Z) id/J/JH/JHI/String-Approx-2.7.tar.gz ... _bigfloatcmp called with l[1237505213.21133]r[UNDEF]: but both must be defined at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/FakeBigFloat.pm line 76 File::Rsync::Mirror::Recentfile::FakeBigFloat::_bigfloatcmp(1237505213.21133, undef) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/FakeBigFloat.pm line 131 File::Rsync::Mirror::Recentfile::FakeBigFloat::_bigfloatlt(1237505213.21133, undef) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/Done.pm line 110 File::Rsync::Mirror::Recentfile::Done::covered('File::Rsync::Mirror::Recentfile::Done=HASH(0x8857fb4)', 1237505213.21133, 0.123456789) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile.pm line 2041 File::Rsync::Mirror::Recentfile::uptodate('File::Rsync::Mirror::Recentfile=HASH(0x8533a2c)') called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recent.pm line 536 File::Rsync::Mirror::Recent::rmirror('File::Rsync::Mirror::Recent=HASH(0x82ef3d0)', 'skip-deletes', 1) called at /home/k/sources/CPAN/GIT/trunk/bin/testing-rmirror.pl line 27 at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/FakeBigFloat.pm line 76 File::Rsync::Mirror::Recentfile::FakeBigFloat::_bigfloatcmp(1237505213.21133, undef) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/FakeBigFloat.pm line 131 File::Rsync::Mirror::Recentfile::FakeBigFloat::_bigfloatlt(1237505213.21133, undef) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/Done.pm line 110 File::Rsync::Mirror::Recentfile::Done::covered('File::Rsync::Mirror::Recentfile::Done=HASH(0x8857fb4)', 1237505213.21133, 0.123456789) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile.pm line 2041 File::Rsync::Mirror::Recentfile::uptodate('File::Rsync::Mirror::Recentfile=HASH(0x8533a2c)') called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recent.pm line 536 File::Rsync::Mirror::Recent::rmirror('File::Rsync::Mirror::Recent=HASH(0x82ef3d0)', 'skip-deletes', 1) called at /home/k/sources/CPAN/GIT/trunk/bin/testing-rmirror.pl line 27 FIXED, it was the "--skip-locking" file where manual intervention was participating * bug on the mirroring slave: when the dirtymark gets increased we probably do not reset the done intervals. The mirrorer stays within tight bounds where it tries to sync with upstream and never seems to finish. In the debugging state file I see lots of identical intervals that do not get collapsed. When I restart the mirrorer it dies with: Sync 1237507989 (227/33111/Z) id/X/XI/XINMING/Catalyst-Plugin-Compress.tar.gz ... _bigfloatcmp called with l[1237400817.94363]r[UNDEF]: but both must be defined at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/FakeBigFloat.pm line 76 File::Rsync::Mirror::Recentfile::FakeBigFloat::_bigfloatcmp(1237400817.94363, undef) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/FakeBigFloat.pm line 101 File::Rsync::Mirror::Recentfile::FakeBigFloat::_bigfloatge(1237400817.94363, undef) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/Done.pm line 226 File::Rsync::Mirror::Recentfile::Done::_register_one('File::Rsync::Mirror::Recentfile::Done=HASH(0x84c6af8)', 'HASH(0xb693f2dc)') called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile/Done.pm line 200 File::Rsync::Mirror::Recentfile::Done::register('File::Rsync::Mirror::Recentfile::Done=HASH(0x84c6af8)', 'ARRAY(0x8c54bfc)', 'ARRAY(0xb67e618c)') called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile.pm line 1044 File::Rsync::Mirror::Recentfile::_mirror_item('File::Rsync::Mirror::Recentfile=HASH(0x84abcb0)', 227, 'ARRAY(0x8c54bfc)', 33110, 'File::Rsync::Mirror::Recentfile::Done=HASH(0x84c6af8)', 'HASH(0x84abd8c)', 'ARRAY(0x839cb64)', 'HASH(0x839c95c)', 'HASH(0xb6a30f0c)', ...) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recentfile.pm line 992 File::Rsync::Mirror::Recentfile::mirror('File::Rsync::Mirror::Recentfile=HASH(0x84abcb0)', 'piecemeal', 1, 'skip-deletes', 1) called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recent.pm line 564 File::Rsync::Mirror::Recent::_rmirror_mirror('File::Rsync::Mirror::Recent=HASH(0x84ab6d4)', 7, 'HASH(0x8499488)') called at /home/k/sources/rersyncrecent/lib/File/Rsync/Mirror/Recent.pm line 532 File::Rsync::Mirror::Recent::rmirror('File::Rsync::Mirror::Recent=HASH(0x84ab6d4)', 'skip-deletes', 1) called at /home/k/sources/CPAN/GIT/trunk/bin/testing-rmirror.pl line 27 and debugging stands at At: before Brfinterval: Z Ci: 227 Dre+1: epoch: 1237400802.5789 path: id/J/JO/JOHND/CHECKSUMS type: new Dre-0: epoch: 1237400807.97514 path: id/M/MI/MIYAGAWA/CHECKSUMS type: new Dre-1: epoch: 1237400817.94363 path: id/X/XI/XINMING/Catalyst-Plugin-Compress.tar.gz type: new Eintervals: - - 900644040 - 900644040 - [] and it is reproducable. Why does the mirrorer not fetch a newer Z file? It is 18 hours old while pause has a fresh one. FIXED, it was the third anded term in each of the ifs in the IV block in _register_one: with that we make sure that we do not stamp on valuable interval data. 2009-03-17 Andreas J. Koenig * done: verified the existence of the floating point bug in bleadperl and verified that switching from YAML::Syck to YAML::XS does not resolve it. BTW, the switch was doable with perl -i~ -pe 's/Syck/XS/g' lib/**/*.pm t/*.t and should be considered as a separate TODO * todo: integrate a dirty update with two aggregate calls before unlocking for frictionless dirtying DONE * todo: start the second rsync daemon on pause DONE * todo: move index files to .recent: this cannot simply be done by setting filenameroot to .recent/RECENT. Other parts of the modules rely on the fact that dirname(recentfile) is the root of the mirrored tree. 2009-03-16 Andreas J. Koenig * What was the resolution of the mirror.pl delete hook bug? Do we call the delete hook when pause removes a file from MUIR? * Today on pause: Updating 2a13fba..29f284d and installing it for /usr/local/perl-5.10.0{,-RC2} TURUGINA/Set-Intersection-0.01.tar.gz was the last upload before this action and G/GW/GWILLIAMS/RDF-Query-2.100_01.tar.gz the first after it 2009-03-15 Andreas J. Koenig * currently recent_events has the side effect of setting dirtymark because it forces a read on the file. That should be transparent, so that the dirtymark call always forces a cache-able(?) read. * The bug below is -- after a lot of trying -- not reproducible on a small script, only in the large test script. The closest to the output below was: #!perl use strict; use Devel::Peek; my $x = "01237123229.8814"; my($l,$r); for ($l,$r) { $_ = "x"x34; } ($l,$r) = ($1,$2) if $x =~ /(.)(.+)/; $r = int $r; $l = "1237123231.22458"; $r = "1237123231.22458"; 1 if $l/1.1; Devel::Peek::Dump $l; Devel::Peek::Dump $r; Devel::Peek::Dump $x = $l <=> $r; die "BROKE" if $x; __END__ The checked in state at c404a85 fails the test with my /usr/local/perl-5.10-uld/bin/perl on 64bit but curiously not with /usr/local/perl-5.10-g/bin/perl. So it seems the behaviour is not even in the test script always consistent. * Todo: write a test that inserts a second dirty file with an already existing timestamp. DONE * Bug in perl 5.10 on my 64bit box: DB<98> Devel::Peek::Dump $l SV = PVMG(0x19e0450) at 0x142a550 REFCNT = 2 FLAGS = (PADMY,NOK,POK,pNOK,pPOK) IV = 0 NV = 1237123231.22458 PV = 0x194ce70 "1237123231.22458"\0 CUR = 16 LEN = 40 DB<99> Devel::Peek::Dump $r SV = PVMG(0x19e0240) at 0x142a3e8 REFCNT = 2 FLAGS = (PADMY,POK,pPOK) IV = 1237123229 NV = 1237123229.8814 PV = 0x19ff900 "1237123231.22458"\0 CUR = 16 LEN = 40 DB<100> Devel::Peek::Dump $l <=> $r SV = IV(0x19ea6e8) at 0x19ea6f0 REFCNT = 1 FLAGS = (PADTMP,IOK,pIOK) IV = -1 DB<101> Devel::Peek::Dump $l SV = PVMG(0x19e0450) at 0x142a550 REFCNT = 2 FLAGS = (PADMY,NOK,POK,pNOK,pPOK) IV = 0 NV = 1237123231.22458 PV = 0x194ce70 "1237123231.22458"\0 CUR = 16 LEN = 40 DB<102> Devel::Peek::Dump $r SV = PVMG(0x19e0240) at 0x142a3e8 REFCNT = 2 FLAGS = (PADMY,NOK,POK,pIOK,pNOK,pPOK) IV = 1237123231 NV = 1237123231.22458 PV = 0x19ff900 "1237123231.22458"\0 CUR = 16 LEN = 40 Retry with uselongdouble gives same effect. Not reproducable on 32bit box (k75). * Todo: reset "done" or "covered" and "minmax" after a dirty operation? DONE 2009-03-11 Andreas J. Koenig * $obj->merge ($other) needs to learn about equal epoch which may happen since dirty_epoch intruded. * Wontfix anytime soon: I think we currently do not support mkdir. Only files! 2009-01-01 Andreas J. Koenig * Todo: continue working on update(...,$dirty_epoch). It must be followed by a fast_aggregate! DONE 2008-12-26 Andreas J. Koenig * maybe we need a closest_entry or fitting_interval or something like that. We want to merge an event into the middle of some recentfile. First we do not know which file, then we do not know where to lock, where to enter the new item, when and where to correct the dirtymark. So my thought is we should first find which file. Another part of my brain answers: what would happen if we would enter the new file into the smallest file just like an ordinary new event, just as an old event? (1) we would write a duplicate timestamp? No, this would be easy to avoid (2) we would make the file large quickly? Yes, but so what? We are changing the dirtymark, so are willing to disturb the downstream hosts. 2008-11-22 Andreas J. Koenig * 10705 root 17 0 725m 710m 1712 S 0.0 46.8 834:56.05 /home/src/perl/repoperls/installed-perls/perl/pVNtS9N/perl-5.8.0@32642/bin/perl -Ilib /home/k/sources/CPAN/GIT/trunk/bin/testing-rmirror.pl leak! https://rt.cpan.org/Ticket/Display.html?id=41199 * bzcat uploads.csv.bz2 | perl -F, -nale '$Seen{$F[-1]}++ and print' Strangest output being HAKANARDO who managed to upload Here is a better oneliner that includes also the first line of each finding: bzcat uploads.csv.bz2 | perl -MYAML::Syck -F, -nale '$F[-1]=~s/\s+\z//; push @{$Seen{$F[-1]}||=[]},$_; END {for my $k (keys %Seen){ delete $Seen{$k} if @{$Seen{$k}}==1; } print YAML::Syck::Dump(\%Seen)}' 2008-10-31 Andreas J. Koenig * memory leak in the syncher? It currently weighs 100M. Update 2008-11-02: root 10705 1.0 4.9 80192 76596 pts/32 S+ Nov02 24:05 /home/src/perl/repoperls/installed-perls/perl/pVNtS9N/perl-5.8.0@32642/bin/perl -Ilib /home/k/sources/CPAN/GIT/trunk/bin/testing-rmirror.pl 2008-10-29 Andreas J. Koenig * lookup by epoch and by path and use this ability on the pause to never again register a file twice that doesn't need it. Let's call it contains(). * after the dirtymark is done: fill up recentfiles with fake (historic) entries; fill up with individual corrections; algorithm maybe to be done with bigfloat so that we can always place something in the middle between two entries. Before we must switch to bigfloat we could try to use Data::Float::nextup to get the. * Inotify2 on an arbitrary tree and then play with that instead of PAUSE directly. * dirtymark now lives in Recentfile, needs to be used in rmirror. * find out why the downloader died after a couple of hours without a net connection. Write a test that survives the not-existence of the other end forever. 2008-10-15 Andreas J. Koenig * reconsider the HTTP epoch only. Not the whole thing over HTTP because it makes less sense with tight coupling for secondary files. But asking the server what the current epoch is might be cheaper on HTTP than on rsync. (Needs to be evaluated) * remove the 0.00 from the verbose overview in the Merged column in the Z row. DONE * write tests that expose the problems of the last few days: cascading client/server roles, tight coupling for secondary RFs, deletes after copies. * Some day we might want to have policy options for the slave: tight/loose/no coupling with upstream for secondary RFs. tight is what we have now. loose would wait until a gap occurs that can be closed. 2008-10-14 Andreas J. Koenig * revisit all $rfs->[$i+1] places if they now make sense still 2008-10-11 Andreas J. Koenig * another bug is the fact that the mirror command deletes files before it unhides the index file, thus confusing downstream slaves. We must not delete before unhiding and must delete after unhiding. FIXED. * new complication about the slave that is playing a server role. Currently we mirror from newest to oldest with a hidden temporary file as index. And when one file is finished, we unhide the index file. Imagine the cascading server/slave is dead for a day. It then starts mirroring again with the freshest thing and unhides the freshest index file when it has worked through it. In that moment it exposes a time hole. Because it now works on the second recentfile which is still hidden. We currently do nothing special to converge after such a drop out. At least not intentionally and robustly and thought through. The algorithm we use to seed the next file needs quite a lot of more robustness than it currently has. Something to do with looking at the merged element of the next rf and when it has dropped off, we seed immediately. And if it ramains dropped off, we seed again, of course. Nope, looking from smaller to larger RFS we look at the merged element of this RF and at the minmax/max element of the next RF. If that $rf[next]->{minmax}{max} >= $rf[this]->{merged}{epoch}, then we can stop seeding it. And we need a public accessor seed and unseed or seeded. But not the mix of public and private stuff that then is used behind the back. And then the secondary* stuff must go. And we must understand what the impact is on the DONE system. Can it go unnoticed that there was a hole? And could the DONE system have decided the hole is covered? This should be testable with three directories where the middle stops working for a while. Done->merge is suspicious, we must stop it from merging non-conflatable neighbors due to broken continuity. FIXED 2008-10-10 Andreas J. Koenig * Slaven suggests to have the current epoch or the whole current recentfile available from the HTTP server and take it away with keepalive. This direction goes the granularity down to subseconds. We might want to rewrite everything to factor out transport and allow the whole thing to run via HTTP. 2008-10-09 Andreas J. Koenig * smoker on k81 fetching from k75 to verify cascading works. See 2008-07-17 in upgradexxx and rsync-over-recentfile-3.pl. * maybe the loop should wait for CHECKSUMS file after every upload. And CPAN.pm needs to deal with timestamps in the future. * do not forget the dirtymark! Text: have a new flag on recentfiles with the meaning: if this changes, you're required to run a full rsync over all the files. The reason why we set it would probably be: some foul happened. we injected files in arbitrary places or didn't inject them although they changed. The content of the flag? Timestamp? The relation between the recentfiles would have to be inheritance from the principal, because any out of band changes would soon later propagate to the next recentfile. By upping the flag often one can easily ruin the slaves. last out of band change? dirtymark? Anyway, this implies that we read a potentially existing recentfile before we write one. And it implies that we have an eventloop that keeps us busy in 2-3 cycles, one for current stuff (tight loop) and one for the recentfiles (cascade when principal has changed), one for the old stuff after a dirtymark change. And it implies that the out-of-band change in any of the recentfiles must have a lock on the principal file and there is the place to set the dirtymark. * start a FAQ, especially quick start guide questions. Also to aid those problematic areas where we have no good solution, like the "links" option to rsync. * wish feedback when we are slow. * reduce mccabe * Remove a few DEBUG statements. * The multiple-rrr way of doing things needs a new option to rmirror, like piecemeal or so. Not urgent because after the first pass through, things run smoothely. It's only ugly during the first pass. * I have the suspicion that the code is broken that decides if the neighboring RF needs to be seeded. I fear when too much time has gone between two calls (in our case more than one hour), it would not seed the neighbor. Of course this will never be noticed, so we need a good test for it. * local/localroot confusion: I currently pass both options but one must do. * accounts for early birds on PAUSE rsync daemon. * hardcoded 20 seconds * who mirrors the index? DOING now. * which CPAN mirrors offer rsync? * visit all XXX, visit all _float places * rename the pathdb stuff, it's too confusing. No idea how. * rrr-inotify, backpan, rrr-register 2008-10-08 Andreas J. Koenig * current bugs: the pathdb seems to get no reset, the seeding of the secondaryttl stuff seems not to have an effect. Have helped myself with a rand(10), need to fix this back. So not checked in. Does the rand thing even help? The rand thing helps. The secondaryttl stuff was in the wrong line, fixed now. The pathdb stuff was because I called either _pathdb or __pathdb on the wrong object. FIXED now. * It's not so beautiful if we never fetch the recentfiles that are not the principal, even if this is correct behaviour. We really do not need them after we have fetched the whole content. OK, we want a switch for that: secondaryttl DONE 2008-10-07 Andreas J. Koenig * bug: rrr-news --max does not count correctly. with "35" it shows me 35 lines but with 36 it shows 110. First it repeats 35, gives 70, and then it lets 40 follow. FIXED * See that the long running process really only updates the principal file unless it has missed a timespan during which something happened. If nothing happened, it must notice even when it misses the timespan. DONE * we must throw away the pathdb when we have reached the end of Z. From that moment we can have a very small pathdb because the only reason for a pathdb is that we know to ignore old records in old files. We won't need this pathdb again before the next full pass over the data is necessary and then we will rebuild it as we go along. DONE 2008-10-06 Andreas J. Koenig * I think, Done::register_one is doing wrong in that it does not conflate neighboring pieces. The covered() method cannot do this because it has no recent_events array at hand. But register_one has it and could do it and for some reason misses to do it (sometimes). This means that the three tests I just wrote can probably not survive because they test with an already broken Done structure. The art now is to detect how it happens, then to reproduce, then write a test, then fix it. So from the logfile this is what happens: we have a good interval with newest file being F1 at T1. Now remotely F1 gets a change and F2 goes on top of it. Locally we now mirror F2 and open a new done interval for it. Then we mirror F1 but this time with the timestamp T1b. And when we then try to close the gap, we do not find T1 but instead something older. We should gladly accept this older piece and this would fix this bug. FIXED * bug to fix: when the 1h file changes while rmirror is running, we do correctly sync the new files but never switch to the 6h file but rather stay in a rather quick loop that fetches the 1h file again and again. Is it possible that we initialize a new object? Or does get_remote_recentfile_as_tempfile overwrite something in myself? Want a new option: _runstatusfile => $file which frequently dumps the state of all recentfiles to a file. FIXED 2008-10-04 Andreas J. Koenig * Todo: now teach update to verify the timestamp is about to write against the previous and use _increase_a_bit if it doesn't comply with strict monotony. DONE * The problem of rounding. So far perl's default precision was sufficient. One day it won't be. FakeFloat has an easy job when it is only reading and other machines have written correctly. But when we want to write a floating point number that is a bit larger than the other one, then we need our own idea of precision. Slaven said: just append a "1". This might be going towards the end of usability too quickly. I'd like something that actually uses the decimal system. Well, appending a 1 also does this but... E.g. we have 1.0. nextup on this architecture is starting with 1.0000000000000004. So there is a gap to fill: 1,2,3. Now I have taken the 1.0000000000000003 and the next user comes and the time tells him 1.0 again. He has to beat my number without stepping over the nextup. This is much less space than I had when I chose 1,2,3. What is also irritating is that nextup is architecture dependent. The 128 bit guy must choose very long numbers to fit in between whereas the other one with 16 bit uses larger steps. But then the algorithm is the same for both, so that would be a nice thing. I see two situation where we need this. One is when Time::HiRes returns us a value that is <= the last entry in our recentfile. In this case (let's call it the end-case) we must fill the region between that number and the next higher native floating point number. The other is when we inject an old file into an old recentfile (we would then also set a new dirtymark). We find the integer value already taken and need a slightly different one (let's call it the middle-case). The difference between the two situations is that the next user will want to find something higher than my number in the end-case and something lower than my number in the middle case. So I suggest we give the function both a value and an upper bound and it calculates us a primitive middle. The upper bound in the middle-case is the next integer. The upper bound on the end-case is the nextup floating point number. But the latter poses another problem: if we have occupied the middle m between x and nextup(x), then the nextup(m) will probably not be the same as nextup(x) because some rounding will take place before the nextup is calculated and when the rounding reaches the nextup(x), we will end up at nextup(nextup(x)). So we really need to consider the nextup and the nextdown from there and then the middle and that's the number we may approach asymptotically. Ugly. But DONE. 2008-10-03 Andreas J. Koenig * consider deprecating the use of RECENT.recent as a symlink. It turns out to need extra hoops with the rsync options and just isn't worth it. Or maybe these extra hoops are needed anyway for the rest of the tree? Nope, can't be the case because not all filesystems support symlinks. But before doing the large step, I'll deprecate the call of get_remote_recentfile_as_tempfile with an argument. Rememberr this was only introduced to resolve RECENT.recent and complicates the routine far beyond what it deserves. DONE. Won't deprecate RECENT.recent, just moved its handling to the supervisor. 2008-10-02 Andreas J. Koenig * I think it's a bug that the rsync_option links must be set to true in order to support RECENT.recent and that nobody cares to set it automatically. Similar for ignore_link_stat_errors. 2008-09-27 Andreas J. Koenig * Todo: find all todos together and make a plan what is missing for a release. - verifytree or something like that. fsck maybe. - rersyncrecent, the script itself? What it do? - a way to only mirror the recentfiles without mirroring the whole remote system such that people can decide to mirror only partially see also 2008-08-30. .shadow-xxx directory? this also needed for a filesystem that is still incomplete and might need the mirrorfiles for lookup(?) - long living objects that mirror again and again. Inject something into ta, see how it goes over to tb. - how do we continue filling up the DONE system when we use an object for the second time? "fully covered" and "uptodate" or new terminology. - overview called on the wrong file should be understandable - the meta data field that must change when we fake something up so that the downstream people know they have to re-fetch everything. - how tolerant are we against missing files upstream? how do we keep track? there are legitimate cases where we did read upstream index right before a file got deleted there and then find that file as new and want it. There are other cases that are not self healing and must be tracked and bugreported. - how, exactly, do we have to deal with deletes? With rsync errors? rsync: link_stat "/id/K/KA/KARMAN/Rose-HTMLx-Form-Related-0.07.meta" (in authors) failed: No such file or directory (2) The file above is a delete in 1h and a new in file 1M and the delete in the locally running rmirror did not get propagated to the 1M object. Bug. And the consequence is a standstill. It seems that a slave that works with a file below the principal needs to merge things all the way up to get rid of later deletes. Or keep track of all deletes and skip them later. So we need a trackdeletes.pm similar to the done.pm? see also 2008-08-20 about spurious deletes that really have no add counterpart and yet they are not wrong. - consider the effect when resyncing the recentfile takes longer than the time per loop. Then we never rsync any file. We need to diagnose that and force an increase of that loop time. But when we later are fast enough again because the net has recovered, then we need to switch back to original parameters. ERm, no, it's enough to keep syncing at least one file before refetching an index file. - remember to verify that no temp files are left lying around and the signal handler - status file for not long running jobs that want to track upstream with a, say, cronjob. - revisit all XXX _float areas and study Sub::Exporter DONE - persistent DB even though we just said we do not need it. Just for extended capabilities and time savings when, for example, upstream announces a reset and we get new recentfiles and could then limit ourselves to a subset of files (those that have a changed epoch) in a first pass and would only then do the loop to verify the rest. Or something. * Todo: aggregate files should know their feed and finding the principal should be done stepwise. (?) * Todo: DESTROY thing that unlocks. Today when I left the debuggerr I left locks around. DONE 2008-09-26 Andreas J. Koenig * maybe extend the _overview so that it always says if and where the last file is in the next file and where the next event in the next rf would lie. No, don't like this anymore. REJECT * take the two new redundant tests out again, only the third must survive. DONE * Todo: add a sanity check if the merged structure is really pointing to a different rf and that this different rf is larger. DONE 2008-09-25 Andreas J. Koenig * now test, if they are overlapping. And test if there is a file in the next rf that would fit into this rf's interval. 1h 1222324012.8474 1222322541.7963 0.4086 6h 1222320411.2760 1222304207.6931 4.5010 missing overlap/gap! 1d 1222320411.2760 1222238750.5071 22.6835 large overlap 1W 1222313218.3626 1221708477.5829 167.9835 I suspect that somebody writes a merged timestamp without having merged and then somebody else relies on it. If aggregate is running, the intervals must not be extravagated, if it is not running, there must not be bounds, the total number of events in the system must be counted and must be controlled throughout the tests. That the test required the additional update was probably nonsense, because aggregate can cut pieces too. FIXED & DONE 2008-09-23 Andreas J. Koenig * rrr-aggregate seems to rewrite the RECENT file even if nothing has changed. FIXED 2008-09-21 Andreas J. Koenig * Most apparent bug at the moment is that the recentfiles are fetched too often. Only the principal should be fetched and if it has not changed, the others should not be refetched. ATM I must admit that I'm happy that we refetch more often than needed because I can more easily fix bugs while the thing is running. * Let's say, 1220474966.19501 is a timestamp of a file that is already done but the done system does not know about it. The reason for the failure is not known and we never reach the status uptodate because of this. We must get over it. Later it turns out that the origin server had a bug somewhere. 1220474966.19042 came after 1220474966.19501. Or better: it was in the array of the recentfile one position above. The bug was my own. 2008-09-20 Andreas J. Koenig * There is the race condition where the server does a delete and the slave does not yet know and then tries to download it because he sees the new. So for this time window we must be more tolerant against failure. If we cannot download a file, we should just skip it and should not retry immediately. The whole system should discover the lost thing later. Keeping track with the DONE system should really be a no brainer. But there is something more: the whole filesystem is a database and the recentfiles are one possible representation of it. It's a pretty useful representation I think that's why I have implemented something around it. But for strictly local operation it has little value. For local operation we would much rather have a database. So we would enter every recentfile reading and every rsync operation and for every file the last state change and what it leads to. Then we would always ignore older records without the efforts involved with recentfiles. The database would have: path,recentepoch,rsyncedon,deletedon Oh well, not yet clear where this leads to. 2008-09-19 Andreas J. Koenig * Bug: the bigloop ran into a funny endless loop after EWILHELM uploaded Module-Build. It *only* rsynced the "1h" recentfile from that moment on. * statusfile, maybe only on demand, alone to have a sharp debugging tool. It is locked and all recentfiles dump themselves into it and we can build a viewer that lets us know where we stand and what's inside. * remember: only the principal recentfile needs expiration, all others shall be expired by principal if it discovers that something has move upstream. 2008-09-18 Andreas J. Koenig * Always check if we stringify to a higher value than in the entry before. DONE * And in covered make an additional check if we would be able to see a numerical difference between the two numbers and if we can't then switch to a different, more expensive algorithm. Do not want to be caught by floating surprises. DONE 2008-09-17 Andreas J. Koenig * caching has several aspects here: we can cache the interval of the recentfile which only will change when the mtime of the file changes. We must re-mirror the recentfile when its ttl has expired. Does have_read tell you anything? It counts nothing at all. Only the mtime is interesting. The ntuple mtime, low-epoch, high-epoch. And as a separate thing the have_mirrored because it is unrelated to the mtime. * Robustness of floating point calculations! I always thought that the string calculated by the origin server for the floating representation of the epoch time is just a string. When we convert it to a number and later back to a string, the other computer might come to a different conclusion. This must not happen, we want to preserve it under any circumstances. I will have to write tests with overlong sequences that get lost in arithmetic and must see if all still works well. DONE But one fragile point remains: if one host considers a>b and the other one considers them == but no eq. To prevent this, we must probably do some extra homework. DONE 2008-09-16 Andreas J. Koenig * the concept of tracking DONE needs an object per recentfile that has something like these methods: do_we_have(xxx), we_have(xxx), do_we_have_all(xxx,yyy), reset() covered() register() covered() The unclear thing is how we translate points in time into intervals. We could pass a reference to the current recent_events array when running we_have(xxx) and let the DONE object iterate over it such that it only has to store a list of intervals that can melt into each other. Ah, even passing the list together with a list of indexes seems feasiable. Or maybe ask for the inverted list? Whenever the complete array is covered by the interval we say we are fully covered and if the recentfile is not expired, we are uptodate. 2008-09-07 Andreas J. Koenig 2008-09-05 Andreas J. Koenig * need a way to "return" the next entry after the end of a list. When the caller says "before" or "after" we would like to know if he could cover that interval/threshold or not because this influences the effect of a newer timestamp of that recentfile. DONE with $opt{info}. 2008-09-04 Andreas J. Koenig * one of the next things to tackle: the equivalent of csync2 -TIXU. loop implies tixu (?). Nope, something like --statefile decides. Per default we do ...? T test, I init, X including removals, U nodirtymark So we have no concept of dirtymarks, we only trust that since we are running we have observed everything steadily. But people will not let this program run forever so we must consider both startup penalty and book keeping for later runs. We keep this for later. For now we write a long running mirror that merges several intervals. 2008-09-02 Andreas J. Koenig * need to speed up the 02 test, it's not clever to sleep so much. Reduce the intervals! * rersyncrecent, the script: default to one week. The name of the switch is --after. Other switches? --loop! 2008-08-30 Andreas J. Koenig * need a switch --skip-deletes (?) * need a switch --enduser that tells us that the whole tempfile discipline is not needed when there is no downstream user. (?) Without this switch we cannot have a reasonable recent.pl that just displays the recent additions. Either we accept to download everything. Or we download temporary files without the typical rsync protocol advantages. Or maybe the switch is --tmpdir? If --tmpdir would mean: do not use File::Temp::tempdir, this might be a win. 2008-08-29 Andreas J. Koenig * apropos missing: we have no push, we never know the downstream servers. People who know their downstream hosts and want to ascertain something will want additional methods we have never thought about, like update or delete a certain file. 2008-08-26 Andreas J. Koenig * tempted to refactor rmirror into resolve_symlink, localize, etc. Curious if rsync_options=links equal 0 vs. 1 will make the expected difference. * rsync options: it's a bit of a pain that we usually need several rsync options, like compress, links, times, checksum and that there is no reasonable default except the original rsync default. I think wee can safely assume that the rsync options are shared between all recentfile instances within one recent tree. 2008-08-20 Andreas J. Koenig * deletes: if a delete follows an add quickly enough it may happen that a downstream mirror did not see the add at all! It seems this needs to be mentioned somewhere. The point here is that even if the downstream is never missing the principal timeframe it may encounter a "delete" that has no complimentary "add" anywhere. 2008-08-19 Andreas J. Koenig * I suspect the treat of metadata is incorrect during read or something. The bug that I am watching is that between 06:08 and 06:09 the 6h file contained more than 6 hours worth of data. At 06:08 we merged into the 1d file. We need to take snapshots of the 6h file over the course of an hour or maybe only between XX:08 and XX:09? Nope, the latter is not enough. Much worse: watching the 1h file: right at the moment (at 06:35) it covers 1218867584-1219120397 which is 70 hours. Something terribly broken. BTW, 1218867584 corresponds to Sat Aug 16 08:19:44 2008, that is when I checked out last time, so it seems to be aggregating and never truncating? No, correct is: it is never truncating; but wrong is: it is aggregating. It does receive a lot of events from time to time from a larger file. Somehow a large file gets merged into the small one and because the "meta/merged" attribute is missing, nobody is paying attention. I believe that I can fix this by making sure that metadata are honoured during read. DONE and test adjusted. 2008-08-17 Andreas J. Koenig * grand renaming plan remotebase => remoteroot to fit well with localroot DONE local_path() => localroot seems to me should already work DONE recentfile_basename => rfilename no need to stress it has no slash DONE filenameroot??? Doesn't seem too bad to me today. Maybe something like kern? It would anyway need a deprecation cycle because it is an important constructor. * I like the portability that Data::Serializer brings us but the price is that some day we might find out that it is slowing us a bit. We'll see. 2008-08-16 Andreas J. Koenig * should we not enter the interval of the principal (or the interval of the merging file?) in every aggregated/merged file? * we should aim at a first release and give up on thinking about sanitizing stuff and zloop. Let's just admit that a full traditional rsync is the only available sanitizer ATM. Otherwise it's complicated stuff: sanitizing on the origin server, sanitizing on the slaves, sanitizing forgotten files, broken timestamps, etc. Let's delay it and get the basics out before this becomes a major cause for mess. 2008-08-13 Andreas Koenig * On OSes not supporting symlinks we expect that RECENT.recent contains the contents of the principal recentfile. Actually this is identical on systems supporting symlinks. Simple, what follows from that is that we need to keep the serializer in the metadata because we cannot read it from the filename, doesn't it? Of course not. It's a chicken and egg problem. This leaves us with the problem to actually parse the serialized data to find out in which format it is. So who can do the 4 or 5 magics we wanted to support? File::LibMagic? 2008-08-09 Andreas Koenig * remotebase and recentfile_basename are ugly names. Now that we need a word for the shortest/principal/driving recentfile too we should do something about it. localroot is good. rfile is good. local_path() is bad, local_path($path) is medium, filenameroot() is bad, remotebase is bad, recentfile is already deprecated. Up to now remotebase was the string that described the remote root directory in rsync notation, like pause.perl.org::authors. And recentfile_basename was "RECENT-1h.yaml". 2008-08-08 Andreas Koenig * The test that was added in today's checkin is a good start for a test of rmirror. We should have more methods in Recent.pm: verify, addmissingfiles. We should verify the current tree, then rmirror it and then verifytree the copy. We could then add some arbitrary file and let it be discovered by addmissingfiles, then rmirror again and then verifytree the copy again. Then we could start stealing from csync2 sqlite database [no port to OSX!] and fill a local DB. And methods to compare the database with the recentfiles. Our strength is that in principle we could maintain state with a single float. We have synced up to 1234567890.123456. If the Z file does not add new files all we have to do is mirror the new ones and delete the goners. This makes it clear that we should extend current protocol and declare that we cheat when we add files too late, just to help the other end keeping track. Ah yes, that's what was meant when zloop was mentioned earlier. Maybe need to revisit File::Mirror to help me with this task. 2008-08-07 Andreas Koenig * There must be an allow-me-to-truncate flag in every recentfile. Without it one could construct a sequence of updates winning the locking battle against the aggregator. Only if an aggregator has managed to merge data over to the next level, truncating can be allowed. DONE with accessor merged. 2008-08-06 Andreas Koenig * We should probably guarantee that no duplicates enter the aggregator array. 2008-08-02 Andreas J. Koenig * To get merge operation faster would need a good benchmark test. What 02 spits out isn't reliable enough and is dominated by many other things. Between commit 10176bf6b79865d4fe9f46e3857a3b8669fa7961 Author: Andreas J. Koenig Date: Sat Aug 2 07:58:04 2008 +0200 and commit 3243120a0c120aaddcd9b1f4db6689ff12ed2523 Author: Andreas J. Koenig Date: Sat Aug 2 11:40:29 2008 +0200 there was a lot of trying but the effect is hardly measurable with current tests. * overhead of connecting seems high. When setting max_files_per_connection to 1 we see that. 2008-08-01 Andreas J. Koenig * 1217622571.0889 - 1217597432.86734 = 25138.2215600014 25138.2215600014/3600 = 6.98283932222261 It jumps into the eye that this is ~ 7 hours, not ~6, so there seems to be a bug in the aggregator. FIXED 2008-07-27 Andreas J. Koenig * e.g. id/Y/YE/YEWENBIN/Emacs-PDE-0.2.16.tar.gz: Do we have it, should we have it, can we mirror it, mirror it! I fear this needs a new class which might be called File::Rsync::Mirror::Recent. It would collect all recentfiles of a kind and treat them as an entity. I realize that a single recentfile may be sufficient for certain tasks and that it is handy for the low level programmer but it is not nice to use. If there is a delete in the 1h file then the 6h file still contains it. Seekers of the best information need to combine at least some of the recentfiles most of the time. There is the place for the Z loop! But the combination is something to collect in a database, isn't it. Did csync2 just harrumph? 2008-07-26 Andreas J. Koenig * it just occurred to me that hosts in the same mirroring pool could help out each other even without rewriting the recentfile. Just fetch the stuff to mirror from several places, bingo. But that's something that should rather live in a separate package or in rsync directly. * cronjobs are unsuited because with ntp they would all come at the full minute and disturb each other. Besides that I'd hate to have a backbone with more than a few seconds latency. 2008-07-25 Andreas J. Koenig * a second rsync server with access control for PAUSE. Port? 873 is the standard port, let's take 8873. * if there were a filesystem based on this, it would have a slow access to inexistent files. It would probably provide wrong readdir (only based on current content) or also a slow one (based on a recentfile written after the call). But it would provide fast access to existing files. Or one would deliberately allow slightly blurred answers based on some sqlite reflection of the recentfiles. * todo: write a variant of mirror() that combines two or more recentfiles and treats them like one * todo: signal handler to remove the tempfile 2008-07-24 Andreas J. Koenig * now that we have the symlink I forgot how it should be used in practice. * the z loop: add missing files to Z file. Just append them (instead of prepending). So one guy prepends something from the Y file from time to time and another guy appends something rather frequently. Collecting pond. When Y merges into Z, things get epoch and the collecting pond gets smaller. What exactly are "missing files"? take note of current epoch of the alpha file, let's call it the recent-ts find all files on disk remove all files registered in the recentworld up to recent-ts remove all files that have been deleted after recent-ts according to recentworld 2008-07-23 Andreas J. Koenig * rersyncrecent might be a cronjob with a (locked) state file which contains things like after and maybe last z sync or such? rrr-mirror might be an alternative name but how would we justify the three Rs when there is no Re-Rsync-Recent? With the --loop parameter it is an endless loop, without it is no loop. At least this is simple. * todo: new accssor z-interval specifies how often the Z file is updated against the filesystem. We probably want no epoch stamp on these entries. And we want to be able to filter the entries (e.g. no by-modules and by-category tree) 2008-07-20 Andreas J. Koenig * Fill the Z file. gc or fsck or both. Somehow we must get the old files into Z. We do not need the other files filled up with filesystem contents though. * need interface to query for a file in order to NOT call update on PAUSE a second time within a short time. 2008-07-19 Andreas J. Koenig * recommended update interval? Makes no sense, is different for different users. * Moosify Local Variables: mode: change-log change-log-default-name: "Todo" tab-width: 2 left-margin: 2 End: