query: fix error reporting for _do_fsck_reset

[MogileFS-Server.git] / CHANGES

2013-01-06: Release version 2.66

   * add a hook to cmd_updateclass (Daniel Frett <daniel.frett@ccci.org>)

   * support updating the class to the default class which has an id of 0 (Daniel Frett <daniel.frett@ccci.org>)

   * reaper: validate DB connection before reaping (Eric Wong <normalperson@yhbt.net>)
     Fixes occasional crash in reaper process.

   * improve handling of classes which change hash algorithm (Eric Wong <normalperson@yhbt.net>)

   * fsck: skip non-existent FIDs properly (Eric Wong <normalperson@yhbt.net>)

   * fsck: use replicate lock when fixing FID (Eric Wong <normalperson@yhbt.net>)

   * query: allow "0" key on all commands which take keys (Eric Wong <normalperson@yhbt.net>)

   * prevent reqid mismatches (and queryworker death) (Eric Wong <normalperson@yhbt.net>)
     Fixes crash case with specific error types.

   * fix use_dest_devs for rebalance (Pyry Hakulinen <pyry@automattic.com>)
     Fixes "use_dest_devs" argument during rebalance.

   * Fix "skip_devcount" during rebalance (Pyry Hakulinen <pyry@automattic.com>)
     Now actually skips updating devcount column during rebalance.

2012-08-13: Release version 2.65

   * Postgres advisory lock instead of table-based lock (Robin H. Johnson <robbat2@gentoo.org>)
     Now requires minimum Postgres version of pg8.4.

   * reaper: switch to Danga::Socket for scheduling (Eric Wong <normalperson@yhbt.net>)

   * reaper: add queue_size_for_reaper server setting (Eric Wong <normalperson@yhbt.net>)

   * reaper: add "queue_rate_for_reaper" server setting (Eric Wong <normalperson@yhbt.net>)
     defaults to 1000, same as previously.

   * reaper: global lock around DB interaction (Eric Wong <normalperson@yhbt.net>)
     prevents reapers clobbering each other, causing a reduction in DB writes.

   * tests: add basic test for reaper (Eric Wong <normalperson@yhbt.net>)

   * fix tests when /etc/mogilefs/mogstored.conf exists (Eric Wong <normalperson@yhbt.net>)

   * iostat: increase flexibility of iostat parser (Eric Wong <normalperson@yhbt.net>)

   * iostat: allow MOG_IOSTAT_CMD env override (Eric Wong <normalperson@yhbt.net>)

   * When a mogstored is down, die with a more informative message. (Dave Lambley <davel@state51.co.uk>)

2012-06-21: Release version 2.64

   * Delete memcache data when we replicate fids (Pyry Hakulinen <pyry@automattic.com>)

   * implement "case_sensitive_list_keys" server setting (Eric Wong <normalperson@yhbt.net>)

   * get_paths: deprioritize devs in "drain" state (Eric Wong <normalperson@yhbt.net>)

   * make marking a host down cause devices to act as down (Eric Wong <normalperson@yhbt.net>)

   * monitor skips hosts marked dead or down (Eric Wong <normalperson@yhbt.net>)

2012-05-29: Release version 2.63

   * Critical bugfix for Postgres users introduced by 2.61. New file uploads
     would fail. (noticed by robin H. Johnson, fixed by Eric Wong)

2012-05-19: Release version 2.62

   * Critical bugfix for a compilation error (dormando, reported by Robin)

   * Fix for upgrading a Postgres install for checksums (Robin * H. Johnson <robbat2@gentoo.org>)

2012-05-18: Release version 2.61

   * fix issue #57 by Pyry and Eric (dormando <dormando@rydia.net>)
     (mogadm host status sometimes allowed typos)

   * avoid unnecessary devcount update in create_close (Eric Wong <normalperson@yhbt.net>)

   * size arg check from tomas doran (dormando <dormando@rydia.net>)
     (pass in a "size" argument (in bytes) to create_open and mogilefs will
      ensure there's enough free space for it)

   * sqlite: implement locking via tables (Eric Wong <normalperson@yhbt.net>)

   * worker/query: Add optional callid parameter (Gernot Vormayr <notti@bgcc.at>)
     (allows command pipelining)

   * delete: prevent orphan files from replicator race (Eric Wong <normalperson@yhbt.net>)

   * fsck: prevent running over 100% completion (Eric Wong <normalperson@yhbt.net>)

   * fsck: cleanup and reduce unnecessary devcount updates (Eric Wong <normalperson@yhbt.net>)

   * fsck: update devcount, forget devs on unfixable FIDs (Eric Wong <normalperson@yhbt.net>)

   * fsck: log bad count correctly instead of policy violation (Eric Wong <normalperson@yhbt.net>)

   * tests: add tests for fsck functionality (Eric Wong <normalperson@yhbt.net>)

   * monitor: only broadcast reject_bad_md5 on change (Eric Wong <normalperson@yhbt.net>)

   * worker: delete_domain returns has_classes error (Eric Wong <normalperson@yhbt.net>)

   * log: enable autoflush for stdout logging (Eric Wong <normalperson@yhbt.net>)

2012-03-30: Release version 2.60

   * Fix fsck status when running for the first time (dormando <dormando@rydia.net>)

   * Checksum support (Major update!) (Eric Wong <normalperson@yhbt.net>)
     See doc/checksums.txt for an overview of how the new checksum system
     works. Also keep an eye on the wiki (http://www.mogilefs.org) for more
     complete documentation in the coming weeks.

2012-02-29: Release version 2.59

   * don't make SQLite error out on lock calls (dormando <dormando@rydia.net>)

   * Postgres wasn't honoring a no-wait timeout (dormando <dormando@rydia.net>)

2012-02-28: Release version 2.58

   * reduce UPDATE's to device table (dormando <dormando@rydia.net>)

   * replicate: retry on short writes to destination (Eric Wong <normalperson@yhbt.net>)

   * enable TCP keepalives for accepted sockets (Eric Wong <normalperson@yhbt.net>)

   * Populate memcache even when noverify is missing (Pyry Hakulinen <pyry@automattic.com>)

   * Remove FID from memcache when it is deleted (Pyry Hakulinen <pyry@automattic.com>)

   * Unset memc when memcache servers have been removed (Pyry Hakulinen <pyry@automattic.com>)

   * Filter out illegal memcache servers from memcache_servers (Pyry Hakulinen <pyry@automattic.com>)

   * Add skip_mkcol setting for servers which do not need it (Pyry Hakulinen <pyry@automattic.com>)

   * skip_devcount variable that disables devcount updates (Pyry Hakulinen <pyry@automattic.com>)

   * Back off if master DB is down (dormando <dormando@rydia.net>)

   * use cached setting for slave list (dormando <dormando@rydia.net>)

   * fix "fsck never stops" issue (dormando <dormando@rydia.net>)

2012-01-31: Release version 2.57

   * Silence a warning about undefined value introduced by the slave database ordering work (hachi@kuiki.net)

   * Need a hook to be able to check the slave list during startup (hachi@kuiki.net)

2012-01-27: Release version 2.56

   * Add database connect timeouts (dormando <dormando@rydia.net>)

   * fix potential hang in JobMaster added in 2.55 (dormando <dormando@rydia.net>)

   * fix "set_weight" command (Eric Wong <normalperson@yhbt.net>)

   * sqlite: consider SQLITE_BUSY and SQLITE_LOCKED deadlocks (Eric Wong <normalperson@yhbt.net>)

   * kill a little more dead fsck/Checker code (dormando <dormando@rydia.net>)

   * Destroy mogdeps (dormando <dormando@rydia.net>)

   * improve slave handling code (dormando <dormando@rydia.net>)

   * don't spew errors if the master goes down (dormando <dormando@rydia.net>)

   * provide an async server setting cache via monitor worker (dormando <dormando@rydia.net>)

   * fix return value of commands that clear_cache (Eric Wong <normalperson@yhbt.net>)

   * make reaper wait less time in test mode (dormando <dormando@rydia.net>)

   * queryworker: reimplement "clear_caches" command (Eric Wong <normalperson@yhbt.net>)

   * worker: use timeout with read_from_parent() instead of sleep (Eric Wong <normalperson@yhbt.net>)

   * Fixed issue where too many devices being active could cause the monitor job
     to be timed out by the watchdog. (Tomas Doran)

2011-11-14: Release version 2.55

   * fixed sources in replication are now a suggestion (dormando <dormando@rydia.net>)

   * More stats to help troubleshoot rebalance (dormando <dormando@rydia.net>)

   * return 1 from condthrow if all is well (dormando <dormando@rydia.net>)

   * schedule fsck replications for the future. (dormando <dormando@rydia.net>)

   * stop enforcing a source id on new file upload (dormando <dormando@rydia.net>)

   * restore new file upload replication priority (dormando <dormando@rydia.net>)

   * Prevent queue related MySQL query hangs (dormando <dormando@rydia.net>)

2011-10-29: Release version 2.54

   * Issue 36: Further insulate from mysql restarts resetting the fid counter (dormando <dormando@rydia.net>)

   * Allow default class to be overridden again. (dormando <dormando@rydia.net>)

   * Fix crash in old delete worker code (Daniel Frett <daniel.frett@ccci.org>)

   * Fix mb_free() counting empty device as missing (git.user <git.user@gmail.com>)

   * sqlite: fix mogdbsetup failure on new setups (Eric Wong <normalperson@yhbt.net>)

   * fid size workaround for lighttpd 1.4.x (dormando <dormando@rydia.net>)

   * Fix various issues in the new mogilefs server query worker. (Tomas Doran <bobtfish@bobtfish.net>)

   * Allow chunked PUT's of unlimited size (dormando <dormando@rydia.net>)

   * Fix listing FIDs.
   * Fix fsck fixing a file from logging a message which appears to be
     two commands (due to spurious \n). Also fix debug function to
     not be able to log messages like this.

2011-08-12: Release version 2.53

   * Fix host status not updating for trackers (important)

2011-07-30: Release version 2.52

   * Plugins (such as FilePaths) might want to be run in info/debug commands (Michael Baehr <mbaehr@plos.org>)

   * schema version 14; support devices > 16TB (dormando <dormando@rydia.net>)

   * Make FSCK worker handle dead devices (dormando <dormando@rydia.net>)

   * Throw out old HTTPFile->size code and use LWP (dormando <dormando@rydia.net>)

   * Remove forced loading of DBD::mysql (dormando <dormando@rydia.net>)

   * Don't test write on readonly devices (Akzhan Abdulin <akzhan.abdulin@gmail.com>)

2011-06-27: Release version 2.51

   * Add minor backcompat for ZoneLocal and friends (dormando <dormando@rydia.net>)

   * Only monitor a device if we should monitor it. (dormando <dormando@rydia.net>)

2011-06-15: Release version 2.50

   * Completely rewire meta object caches. Device/Host/Class/Domain caches are
     now asynchronously loaded via the Monitor worker. This should drastically
     reduce the number of small queries issued against the master database.
     (dormando <dormando@rydia.net>)

   * Don't run workers until Monitor runs once (dormando <dormando@rydia.net>)

   * s/TYPE/ENGINE/ for innodb initialization (dormando <dormando@rydia.net>)

   * Create mogile database explicitly with UTF-8 encoding. (Robin H. Johnson <robbat2@gentoo.org>)

   * Actually use the MOGTEST_* variables for Postgres testing... (Robin H. Johnson <robbat2@gentoo.org>)

   * Add support for config file includes for Mogstored under lighttpd (Jason Mills <jmmills@cpan.org>)

   * Make postgres use new delete worker code (Robin H. Johnson <robbat2@gentoo.org>)

   * New test env variables: MOGTEST_DBROOT{USER,PASS} for easier test usage. (Robin H. Johnson <robbat2@gentoo.org>)

   * Document test variables. (Robin H. Johnson <robbat2@gentoo.org>)
   
   * Support inclusion of custom lighttpd config. (Jason Mills <jmmills@cpan.org>)

2011-01-13: Release version 2.46

   * Fix job_master crash under postgres (Robin H. Johnson <robbat2@gentoo.org>)

   * Revert "only dbh->ping if not used in over a minute" (dormando <dormando@rydia.net>)
      *IMPORTANT FIX* if you're using 2.45, you should upgrade to this.

   * Example application. (dormando <dormando@rydia.net>)

   * missing $fidid for source_down error message (dormando <dormando@rydia.net>)

2011-01-08: Release version 2.45

   * file_debug command (dormando <dormando@rydia.net>)

   * add "file_info" command for fetching metadata. (dormando <dormando@rydia.net>)

   * don't allow clients to upload to the wrong dev (dormando <dormando@rydia.net>)

   * validate database handles in the right places (dormando <dormando@rydia.net>)

   * only dbh->ping if not used in over a minute (dormando <dormando@rydia.net>)

   * make list_fids work with gaps (dormando <dormando@rydia.net>)

   * Postgres enqueue_fids_* cleanup (Robin H. Johnson <robbat2@gentoo.org>)

   * Fixes MogileFS::Store::retry_on_deadlock to honor non-deadlock exceptions (Jason Mills <jmmills@cpan.org>)

   * Implements tests for MogileFS::Store::retry_on_deadlock (Jason Mills <jmmills@cpan.org>)

   * Implements tests for MogileFS::Store::ignore_replace (Jason Mills <jmmills@cpan.org>)

   * job_master was never revalidating its dbh handle. (dormando <dormando@rydia.net>)

   * add a couple useful stats (dormando <dormando@rydia.net>)

   * Correct iostat command on darwin (Tomas Doran <bobtfish@bobtfish.net>)

   * Fix test db creation with SQLite (Tomas Doran <bobtfish@bobtfish.net>)

   * Fix SQLite so tracker at least starts (Tomas Doran <bobtfish@bobtfish.net>)

   * additional debian packaging fixes (kad <kad@blegh.net>)

   * debian packaging updates. (Andre Bohr <andrepascha@googlemail.com>)

   * Documentation updates. (Pyry Hakulinen <pyh@iki.fi>)

   * HTTPFile returns -1 if the file is missing from the device. Content-Length: 0 is valid and it should return BLEN not MISS. (Pyry Hakulinen <pyh@iki.fi>)

   * some further editing of the info file. (dormando <dormando@rydia.net>)

   * Changed repository metadata (Tomas Doran <bobtfish@bobtfish.net>)

2010-11-25: Release version 2.44

        * Fix issue with rebalance not filtering source devs from the
          destination list.

        * Make global rebalance limits work (Martijn Lina)

2010-10-10: Release version 2.43

        * Make over-replication fixable again. (sigh)

        * Fix old bug in MultipleHosts() which would not let you drop from
          mindev > 1 down to 1.

2010-10-08: Release version 2.42

        * Make FSCK run again (Andre Pascha)

        * Fix Postgres schema upgrade issue (hopefully)

2010-10-06: Release version 2.41

        * One line change that to fix uprading to 2.40

2010-09-28: Release version 2.40

        * Optimize worker sleep patterns and worker/parent communications.

        * New rebalance/drain code. See the docs/wiki for how to use.
          ***NOTE*** Old drain/rebalance code is disabled. Setting a device
          into drain mode no longer does what you think it would.
          http://code.google.com/p/mogilefs/wiki/Rebalance

2010-08-13: Release version 2.37

        * Make HTTP DELETE's work reliably

        * Fix obscure crashes related to "unknown devices"

        * make it slightly harder to crash trackers with bad slave DB settings
          (Tomas Doran)

        * make tracker -> storage node connection timeout configurable.
          was hardcoded to 2 seconds. (Tomas Doran)

        * update_class command (Tomas Doran)

2010-04-16: Release version 2.36

        * Bump required version of perlbal for mogstored.

        * Critical bug fix for creating files.
          If you have many dead devices, poor filter ordering could cause
          MogileFS to not consider any alive devices for new files.

        * Actually default to BIGINT under MySQL...
          Bug would always initialize new setups to INT.

2010-04-02: Release version 2.35

        * Nuke ancient mogilefs 1.1 replication code.
          no_unreachable_tracking and old_repl_compat variables are gone.
          (dormando)

        * Support modifying class replpolicy strings (requires new
          mogadm/MogileFS::Client) (dormando)

        * Print fidid and devid on errors when fsck fails to fix something
          (t0m)

        * Allow list_keys to work with a prefix of '0' (dormando)

        * Print timestamp when logging to STDOUT (t0m)

        * Sort by free space and weight fewer devices to avoid CPU churn
          (dormando)

        * Fix deadlock-related crash bugs in db queue fetch routines.

        * Fix dependencies in MogileFS::Server (t0m)

        * Fix bug in list_keys command that prevented searches with a
          '0' prefix. (Ask Bjørn Hansen)

2009-12-05: Release version 2.34

        * Fix bug in new copycount reduction code. If it was more than off by
          one it wouldn't reduce the count properly.

        * Improve accept loop. A reasonably busy tracker could start dropping
          new client connections too easily.

2009-11-24: Release version 2.33

        * 'mogadm fsck status' will now be more accurate.

        * Support non-GNU df (paul armstrong)

        * Remove old !replication command. It's in `mogadm stats` and tends
          to kill the parent process.

        * Force refreshing internal monitoring status every 5 minutes. Help
          work around a potential bug where state changes are missed.

        * Use new delete queue for tempfile culling

        * Calculate free space from available space, not total space
          (todd lipcon)

        * FSCK code has new tunables to control how aggressive it runs:
          queue_size_for_fsck (how many fids to keep in file_to_queue table)
          queue_rate_for_fsck (how many fids to put into the table per second)
          the generic internal queue is now tunable:
          internal_queue_limit (how many fids to pull out of a queue table at
              once. old default was 500).
          All are tunable by 'mogadm settings set'
          (dormando)

        * Replication will now attempt to handle over replication situations.
          FSCK will catch over replication as a policy violation, and
          replicate will attempt to fix it. (dormando)
          (rarely it might need multiple fscks)

        * Since FIDSizes is no longer used, remove it and remove gearman
          dependency. (dormando)

        * FSCK no longer rabidly fills the queue table. (dormando)

        * Make storage code resilient to rare deadlocks (dormando)

        * Add 'max_handles' config option to restart a DB connection if there
          are too many handles to it (victori).
 
2009-10-18: Release version 2.32

        * No changes... MFS is hard to release due to mogdeps :/

2009-10-18: Release version 2.31

        * Fix create_close race condition. (pipegeek)

        * Fix bug in reporting of available disk space.

        * Many minor bug/documentation fixes.

        * MySQL and Postgres installs now use BIGINT's for fid/length by
          default. (dormando, robbat2)

        * Accept 405 responses to MKCOL from the WebDAV server since it often
          just means that the directory already exists. (Ask Bjørn Hansen)

        * Fix showing domain and class in mogadm output when there's only the
          default class in the domain (Mark Imbriaco)

        * mogstored: pidfile support

        * New FSCK work and error for BCNT for bad devcount values (they would
          not otherwise be caught unless the triggered another fix).

2009-01-12: Release version 2.30

        * fsck will now fix broken 'devcount' cols.

        * schema version bump to 12, and new delete worker code to use the new
          queue system. delete should be faster and more scalable.

        * add server setting queue_rate_for_fsck to allow rate limiting how
          fast the new distributed fsck creates jobs. the original default of
          10k could utterly defeat wimpy databases.

        * Support 0 byte files.

        * Fix a bug where rebalance would accidentally delete copies of a fid.

        * Create new internal queue management system, with new job
          'JobMaster'. This greatly reduces database load and increases
          throughput for workers using the new system.

        * Add 'no_unreachable_tracking' config. Disables more of the old
          replication code. Use this if you have no 1.x trackers.

        * Add hook for alternate replication sources. Obscure, but can be
          nice.

        * Add replicate_order_final_choices hook, so multidc replication may
          pull from the local network if one copy already exists.

        * Fix subtle bugs in process management.

        * Allow overriding of default classes.

2008-08-18: Release version 2.20

        * One more time, and hopefully never again. No changes. META.yml
          fixes.

2008-08-18: Release version 2.19

        * No changes except for META.yml fixes.
          _should_ be able to upload releases without owning half of CPAN.

2008-08-16: Release version 2.18

        * Add repl_use_get_port option (Brian Lynch)

        * Remove tempfile data and rows on create_close (Chaos Wang)

        * Don't delete files accidentally during tempfile table cleanup

        * Specify username to run the mogstored daemon (Jonathan Share)

        * Many spelling errors corrected.

        * PostGres locking routines did not include the hostname for the lock
          release.

        * RFC 2518 says we "should" use a trailing slash when calling
          MKCOL. Some servers (nginx) appears to require it. (Spotted by
          Timu Eren).

        * Add server option 'rebalance_ignore_missing' in mogilefsd config
          that allows the rebalance to ignore files that are already missing
          when it tries to delete them.

        * Fix MultipleHosts replication policy to no longer return desperate
          devices after no longer being desperate, instead returning a state
          which will queue for retries in the future till it can be truly
          happy. This prevents files from being replicated to all writable
          devs when there are two few writable hosts to satisfy the policy.

        * Add --chmod-mountpoints option to mogautomount, in an attempt to
          prevent mogstored machines from accidentally writing to their under-
          lying filesystem.

        * New optional argument 'startpos' to fsck_reset.

        * Use correct per-Store port, defaulting to UNIX sockets on localhost.

        * add --dbport option to mogdbsetup (Ask Bjoern Hansen)

        * Fix state_change() bug that could make file_to_delete_later
          grow unnecessarily (Radu Greab)

        * Add Mogstored::HTTPServer::None to allow using mogstored
          with unmanaged backend DAV servers.

        * Add experimental cmd_edit_file. Client requires DAV MOVE and
          partial PUT on the store to use it.

        * make mogstored and mogilefsd use the now-bundled dependency
          Perl modules as a last resort, unless MOGILE_NO_BUILTIN_DEPS
          environment variable is true.  this makes it easier for people
          to use mogilefs, without dependency hell.  distros/packagers
          are expected to ignore (or scoff at) this "feature".  but
          easier for people working from tarball/svn releases.  bundled
          dependencies will be updated over time, as needed.

        * Bring the SQLite store up to date

        * Increase the timeouts in the test suite

        * In the test suite, also search for mogadm in /usr/local/bin/ and
          /usr/local/sbin (spotted by Dean Wilson)

        * SCHEMA VERSION 10:  'replpolicy' column on 'class' table.
          it's safe to --ignore-schema-version and run this mogilefsd
          against an older schema.  new column enables per-class
          pluggable replication policies.  it was always
          abstract, but not easy to plugin your own alternatives.
          see doc/pluggable-replication-policies.txt

        * add start of a MogileFS::ReplicationPolicy::Union implementation

        * fix crash in queryworker's create_open command, when a device
          has its directory made on a mogile storage node (the host of
          which is being monitored by the 'monitor' process), but the
          device isn't registered in the devices table yet.

2007-07-03: Release version 2.17

        * get_paths command now takes a pathcount option, defaulting to
          2, instead of having a hardcoded constant of 2. This allows
          requesting more than 2 paths to a file.

        * make fsck log over-replicated files as policy violations.  it
          won't delete over-replicated files, though, just log

        * Document how to test MogileFS under doc/testing.txt. You really
          should read this if you want to use MogileFS to store your data.

        * Add a safety check to ensure that Postgresql-8.2 or newer is used.

        * Update mogdbsetup usage, and ensure it chooses the correct admin
          username automatically.

        * Document the --type option to mogdbsetup.

        * Introduce the PostGres Store implementation. The locking system might
          still have issues, beware of stale locks! Initial prototype by Radu
          Greab <rg@yx.ro>, and developed heavily by Robin H. Johnson
          <robbat2@gentoo.org>.

        * start of support to make mogstored wrap Apache + web_dav

        * Remove duplicated [fsck] in log output of Fsck worker.

        * Handle HiRes gettimeofday correctly to get useful results. Arrays are
          important. This gives us correct timing output.

        * Handle input to log() correctly. Based on the implementation of
          Sys::Syslog::syslog(), produces no ugly warnings, and makes sure that
          printf stuff works nicely.

        * The unix_timestamp function belongs to the Store, not other classes.
          Need to add test cases for process_deletes and stats.

        * Using NULL during an INSERT for the AUTO_INCREMENT column does not
          work very well outside of MySQL. Pass the correct number of arguments
          to the INSERT instead.

        * Clean up all raw usages of UNIX_TIMESTAMP() function, in preparation
          for database independence. If you need the unix timestamp, include the output
          of $self->unix_timestamp directly into your query.

        * Alter $0 so that the scripts show up as '$SCRIPTNAME' in the process
          listing instead of '/usr/bin/perl $SCRIPTNAME'. This makes init.d
          scripts easier to manage.

        * make mogstored's lighttpd config use mod_status for "/", so
          mogadm doesn't think it's dead when it previously saw a 404
          (Justin Huff <jjhuff@mspin.net>)

        * remove Linux::AIO support, now that Perlbal has removed it as well,
          and because Linux::AIO support (when Perlbal then chooses
          IO::AIO) was hanging mogstored on shutdown.  so also add a
          mogstored shutdown test (after a full daemonization of it)

        * make Mgd::validate_dbh() re-check/ping slave dbhs as well.
          (problem mentioned by Steven Shou <steven.shou@verizon.net>)

        * support drive identifiers above sdz (Arjan <info@bouncetek.com>)

        * make mogstored's iostat child process properly shutdown,
          even when iostat binary isn't installed.  (previously
          it was stuck in a loop every 10 seconds, hoping for
          iostat to be installed, but never respecting parent
          process' death)

        * Change delete worker to use new DeviceStates interface.

        * Fix get_paths to honor new device state of 'drain' via new
          DeviceStates interface.

        * use significantly less CPU when re-broadcasting around
          iostat device utilization% to child processes.  (future:
          don't rebroadcast at all and use shared memory)

        * slightly faster fsck (do less syscalls)

        * efficiency: don't broadcast dup state information to children
          (track in parent what children know already)

        * if using Gearman::Server >= 1.09, respawn the fidsizes
          worker if it dies.

2007-05-09: Release version 2.16

        * BUG FIX from 2.15:  partially written but never closed
          tempfiles weren't being cleaned up properly.

        * start to work better when compiled as a stand-alone
          PAR all-in-one script/archive

2007-05-07: Release version 2.15 ("fsck/drain/rebalance")

        * minor bug fixes and enhancements for MultipleHosts replication
          policy.  bunch of corner cases now checked with new tests.
          now easy to write more replication policy tests in future.

        * replication policies can now return "desperate" requests,
          signalling that a replication reassessment should be
          enqueued for the future, to see if things could be improved.
          (that part's not currently implemented, but the real feature
           and motivation is that the rebalancer now won't delete a
           DevFID if it results in a desperate move, only an ideal move. )

        * replication policies now can optionally return a new return
          value of the (new) type MogileFS::ReplicationRequest, which
          has pretty accessor names, can suggest multiple places,
          can indicate non-ideal emergency replication decisions.
          old plugins' return values will be transparently upgraded
          to the equivalent new return value objects.

        * adding new device state: "drain".  it's a hybrid of "dead"
          (in that files are migrated off it) and "readonly", it that
          it still serves traffic... it just doesn't get new files.
          this also introduces the new object-oriented DeviceState class,
          and device_state($name) utility function to get the DeviceState
          singleton by name

        * internal code cleanup.  notably, kill the old & nasty legacy
          'find_deviceid' function which was ridiculously long and hairy.
          the two callers are now more readable with sorts/greps/etc.

        * make mogstored's devN/usage writing process (DiskUsage) be less
          racy with the mogilefsd monitoring code... don't open file for
          write... open read/write, then in one write system call, write
          the entire file, with newline padding at end to cover old data,
          then truncate it if necessary.  should remove harmless (but scary)
          error messages previously reported by the mogilefsd monitor
          about zero-length usage files.

        * new protocol commands to list/set (certain) server settings,
          with value sanity checking (see MogileFS::Config for which
          are settable, and with what values).  needed for "enable_rebalance".
          was partially enabled before for slave settings.  also needed
          for memcached support before, which was never possible to
          set with mogadm, only with db tweaking.

        * make mogilefsd fsck use new mogstored fid_sizes command, to
          do bulk stats.  speeds up fscks a ton.

        * be robust against system clocks that go backwards between
          gettimeofday calls:
          http://lists.danga.com/pipermail/mogilefs/2007-April/000897.html

        * Put gearman server in mogstored process, add worker
          'mogstored-fidsizes' which runs as subprocess of mogstored. Add
          side-channel command 'fid_sizes' which allows us to quickly enumerate
          and get sizes for files across entire devices on a storage node.

        * remove all code like $state eq "readonly", $state=~ /^dead|down$/
          and instead convert it into specific questions on policy/traits
          of given state, like $dev->should_put_new_files_on, or
          $dev->should_drain_files_off.  see MogileFS::DeviceState,
          objects of which are accessed via $dev->dstate, or new
          MogileFS::Util device_state($name) wrapper.

        * start of rebalance support.  (where replication workers, in their
          idle time, can rearrange files to even out disk space and/or IO
          activity on storage nodes... policy isn't hard-coded, and is
          in fact currently random)

        * lighttpd support in both mogilefsd and mogstored.  passes test
          suite with environment MOGSTORED_SERVER_TYPE=lighttpd set now.

        * abstract out the HTTP server support in mogstored, so
          mogstored isn't just a perlbal wrapper, but an anything
          wrapper. (in particular, lighttpd and apache)  mogstored still
          exists for all its other misc admin/monitoring functions,
          but can then manage/configure apache/lighttpd child process(es).
          so far they're just stubbed out.

        * split mogstored into separate files per class, rather than one
          large script.

2007-04-24: Release version 2.10

        * bug from earlier refactoring:  changing some 'next' to 'return'
          in MogileFS::Device's create_directory method.  (thanks to
          Justin Huff <jjhuff@mspin.net>)

        * make sure mogdbsetup doesn't upgrade an existing table (with
          potentially lot of data) from MyISAM to InnoDB.  just warn.
          only new tables should become InnoDB on creation.  also,
          detect InnoDB even on old MySQL 4.0.x, not just 4.1/5.0.

        * make fsck faster by reducing number of db queries (both
          reads and writes), though more unrelated fastness remains to
          be done.

        * make fsck_status faster (keep summary stats of evcode counts,
          rather than doing a full GROUP BY query all the time)

2007-04-20: Release version 2.09

        * filesystem checker (Fsck worker).  off by default, unless
          it's enabled via protocol.  Control/monitor it with mogadm
          in MogileFS-Utils 2.08, which needs MogileFS::Client 1.06.

        * replication speed-up on configurations with only 1 host.
          (replication would randomly have no suggestions before, when
           there was only 1 target host, so replication would be rescheduled
           until the future, when random returned a different device on
           that host)

2007-04-17: Release version 2.08

        * (no runtime-visible changes from 2.07)

        * mogilefsd: remove incomplete "Checker" worker, replace with
          incomplete "Fsck" worker.

        * test cleanup/robustness improvements.  properly skips testing when
          testing deps not found.  but searches harder for them now, too.
          tests run under "make disttest".

        * packaging cleanups, released using ShipIt.


2007-03/04:

        * mogilefsd: Split Plugin::MetaData and Plugin::FilePaths into their own distributions.

        * mogilefsd: Fix mogdbsetup to not ask empty questions.

        * mogilefsd: command line/config file support for listening on specific IPs, also
           support for listening on multiple IPs/ports

2007-03-19:
        * mogstored: use XS HTTPHeaders by default, unless disabled with environment variable

        * mogilefsd: memcache support for get_paths (see doc/memcache-support.txt)

2007-03-16:
        * mogstored: make child processes be exec'ed (as as their own scripts, not inheriting
          Perlbal, Linux::AIO/IO::AIO, etc baggage).  so they don't own listening
          sockets, pipes to child workers, signal handlers, etc.  just a lot
          cleaner now, and SIGTERM, SIGINT properly shut everything down now,
          both in foreground and as daemon

2007-03-15:
        * mogstored: this changelog is totally neglected.  :/  see svn log.

        * mogstored: make perlbal's "shutdown graceful" work, by not doing
          sidechannel accepting ourselves (which never hung up),
          but subclass Perlbal's TCPListener class, which does it
          all correctly.

2006-12/2007-01:  (2.00_05)
        * mogilefsd: everything converted to OO API internally.  old internal APIs removed.

        * mogilefsd: be database-agnostic.  Currently only works with MySQL (and
           SQLite as a joke/demo), but Pg/Oracle coming in the future.

        * mogilefsd: built-in I/O activity monitoring between mogstored/mogilefsd Monitor
           now.

        * mogilefsd: much improved internal communication between worker processes

        * mogilefsd: start of better docs

2006-12-28
        * mogilefsd: pidfile support (Radu Greab <radu@yx.ro>)

2006-12-28
2006-12-27
        * mogilefsd: remove all old internal APIs, update everything to use new OO APIs,
           expanding OO APIs where necessary (lots of places).  so much more
           beautifully hackable now.

        * mogstored: kill child processes (diskusage and iostat) when being killed.

2006-11-05:
        * mogstored: set aio threads after daemonizing.  needed by Linux::AIO/IO::AIO, it seems?

2006-09-29
        * mogilefsd: remove non-HTTP mode ("NFS mode") from the server

2006-09-23:
        * mogstored: refuse to start up with missing Linux::AIO/IO::AIO or broken Linux::AIO

2006-09-04:
        * mogstored: set aio_threads as a function of number of disks

2006-07 -> 2006-09
        * mogilefsd: major refactoring/cleanup/new features ("MogileFS 2.0")

2006-03-01 (Brad)
        * mogilefsd: fix dataloss bug when the 'tempfile' table is InnoDB and the
           server is restarted while the tempfile table is empty.

2006-02-28 (Jr)
        * mogilefsd: allow create_open to take optional fid parameter which enables you to
        dictate which fid to use internally and to not rely on MySQL's auto_increment

2006-01-10 (Brad)
        * mogilefsd: make Mgd::get_device_summary() always return all devices, regardless
        of status.

        * mogilefsd: update all callers of get_device_summary to filter on the device
        status, if they need to. (provided more clarity into what
        callers' intentions were...)

        * mogilefsd: add a new device state 'readonly' which means: it can be read from
        from clients, it can be replicated from, but no new files can go
        on it or be replicated onto it. and it also reaps, meaning the
        device is fucked. the filesystem probably ate itself.

2005-08-10:
        * mogstored: write out usage files for devices, even if they're not on their own mount point.
          also explicitly include the '-k' flag to df, so we know they're KiB.

2005-08-07:
        * mogstored: new config syntax and command line options