bk commit into 5.1 tree (jonas:1.2044)

[jonas]

Below is the list of changes that have just been committed into a local
5.1 repository of jonas. When jonas does a push these changes will
be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html

ChangeSet@stripped, 2006-09-07 11:32:59+02:00, jonas@stripped +1 -0
  Add changelog to alcatel clone'

  cluster_change_hist.txt@stripped, 2006-09-07 11:31:38+02:00, jonas@stripped +423
-0
    BitKeeper file /home/jonas/src/mysql-5.1-wl2325-5.0/cluster_change_hist.txt

  cluster_change_hist.txt@stripped, 2006-09-07 11:31:38+02:00, jonas@stripped +0 -0

# This is a BitKeeper patch.  What follows are the unified diffs for the
# set of deltas contained in the patch.  The rest of the patch, the part
# that BitKeeper cares about, is below these diffs.
# User:	jonas
# Host:	perch.ndb.mysql.com
# Root:	/home/jonas/src/mysql-5.1-wl2325-5.0
--- New file ---
+++ cluster_change_hist.txt	06/09/07 11:31:38
(Version 1.10)

Complete list of bugfixes (related to Cluster) from Alcatel drop 1 (25 Jan 2005).

Note that the names for the bugs are most often given by the users reporting them,
thus they very often are not very descriptive to what the actual bug is.
More information on the bugs, features and a comprehensive list of all MySQL bugs fixed
per release
can be found on the MySQL website http://dev.mysql.com/doc/mysql/en/news.html.
(note that this does not apply to the not yet released Alcatel features for Partitioning
and Replication)
(* == Alcatel has hit this bug)


Alcatel Drop 5p14 (22 May 2006)

Bugs fixed:
   (Bug #19930) Incorrect thread handling in ndb_mgmd could casue crash/incorrect behavior
   (Bug #18591) Repeated "show" and "all status" calls to ndb_mgm causes mgmd to crash
  *(Bug #13987) Cluster: Loss of data nodes can cause high CPU usage from ndb_mgmd


Alcatel Drop 5p13 (8 May 2006)

Functionality added or changed:
   remove TransporterFacade singleton to allow connections to several clusters from same
process

Bugs fixed:
   (Bug #17150) Cluster doesn't allow max number of connections
   (no bug report) removed leak of ndb event operations on delete of ndb object
  *(Bug #19069) replication: does not allow replication to slave with different number of
columns
        - partial solution, allowing more columns on the slave
   (no bug report) replication: Make sure that SUB_GCP_COMPLETE_ACK is sent uncond,
failure to do so can lead to problems...
   (no bug report) Send TC_COMMIT_ACK uncond, failure to do so can lead to problems...
  *(Bug #18550) ndbd getting "node failure handling not complete..." after graceful
restart
      - addded more retries to wait for nodefailure to complete
   (Bug #19039) multi node failure causes node failure handling not to complete
      - patch to avoid this scenario when the management server is used to perform the
stop
      - wait for NF_COMPLETE_REP in management server before returning
   (no bug report) allocate nodeid - only retry on retryable error to give less printouts
in cluster log and faster detection of "bad" connect
   (Bug# 18597) Handle early abort of scan, so that resources are correctly freed
  *(Bug #18116) MySQL 5.0 deadlocks while writing binary log
   (no bug report) replication: stilled mem leak in find all tables


Alcatel Drop 5p12 (10 Apr 2006)

Functionality added or changed:
   Support for ndb multi-node shutdown

Bugs fixed:
   (Bug #18932) Cluster binlog mysqld accepts updating although binlog not setup
  *(Bug#15695 Bug#16447 Bug#18612) For various reasons have a partitioned cluster been
created
      1) make sure it's detected
      2) shutdown is forced if detected
      3) add two new start options that allows control of partial start to avoid the
situation


Alcatel Drop 5p11 (30 Mar 2006)

Bugs fixed:
   (Bug #6061) mysql-log-rotate script - error logging doesn't use new file
   (Bug #18614) ndb: mysqld hangs in alter table
  *(Bug #18491)	cluster: node restart with pending dropeed events causes failed restart
   (Bug #18385) partial restart failure after create table
   (Bug #16772) Starting node joins cluster too early
   (Bug #18352) Rollback + NF can lead to stray locks
   (Bug #18298) 8 Node restart with table/index wo/ logging cause crash
   (Bug #18118) Master node shutdown caused full cluster crash
  *(Bug #18451) Restoring a backup changes fragement type


Alcatel Drop 5p10 (20 Mar 2006)

   (Bug #18349) Wrong error message "Array index out of bounds..."
   (Bug #14028) [Data Loss] NDB node crashes during backup if a BLOB is inserted and
deleted
  *(Bug #18249) alter table on partitioned cluster table fails


Alcatel Drop 5p9 (16 Feb 2006)

Bugs fixed:
  *(Bug#17459) Cluster row based replication not setup if backup is ongoing
  *(Bug#10987) After NDBD failure, restart NDBD dies stating Unable to find restorable
replica
  *(Bug#17295) Unable to read REDO log


Alcatel Drop 5p8 (1 Feb 2006)

Upgrade notes:
   Online upgrade requires ndbd nodes to be upgraded before management server
   No application recompile required

Functionality added or changed:
  *multi chunk feature, new config parameter DataMem2
  *new config parameter to specify number of filesyestem threads to start at startup,
InitialNoOfOpenFiles
   added debug printout to collect more info on unable to find restorable replica

Bugs fixed:
  *(Bug#16996) ndbd nodes crash at api heartbeat failure if event/replication is used
  *(Bug#16772) Starting node joins cluster too early, workaround, do not hand out nodeid
from management server


Alcatel Drop 5p7 (18 Jan 2006)

Upgrade notes:
   Bugfixes in node restart protocol, reguires complete cluster restart.  Discussed and
ok'd by Alcatel/Gerard on the 18:th.
   Some bugfixes require that the API's be recompiled to take effect, but they will
function without recompiling.

Functionality added or changed:
  *added two methods to retrieve information about an NdbDictionary::Event
  *added timeout functionality to management server to avoid having to use "purge stale
sessions"

Bugs fixed:
  Selected bugfixes up until 5.0.19 as listed below
  (Bug#16152) create event assertion in debug version, list corruption
  (Bug#16125) Bitfield with offset+len = 32 crashes cluster (debug compiled)
  (Bug#15908) NF of starting node during NR, when scan is complete on primary yields error
  (Bug#15878) Erroneous error message, Array index out of range
  (Bug#15685) Incorrect abort handling in case of timeout during abort
  (Bug#15682) Incorrect handling of varchar primary keys, when doing filesort
  (Bug#15632) Race condition in INCL_NODEREQ and GCP_PREPARE during NR to starting node
  (Bug#15619) Test case 'ndb_autodiscover' failure on Linux/Itanium
  (Bug#15587) NF during CopyFragReq can lead to repeated NF during NR
  (Bug#15512) crash during online add of API/SQL node
  (Bug#15425) Small window for NF during backup failing without error
  (Bug#15215) mysqld fails at start-up because of illegal reply from mgmd
  (Bug#14199) Leak of ConcurrentIndexOperations
  (Bug#13078) mysqld memleak for ndb


Alcatel Drop 5p6 (30 Nov 2005)

Functionality added or changed:
    *added experimental option to ndbd to use sbrk and trace memory allocation

Bugs fixed:
    *moved initialization of ndbd fs block first to ensure that it gets enough space for
allocation of file system thread stacks
    *make sure ndb threads are not started with too small stacksize (which results in
default, much too big, stack size to be used)
    *fix error handling if thread creation fails in ndbd
    *(Bug#14971) Purge master logs command may hang mysql server for cluster replication
    *correct error message in ndb_restore


Alcatel Drop 5p5 (1 Nov 2005)

Upgrade Notes:
    Upgrading from previous versions should be done like follows:
    - Take appropriate backups, note that table definitions a'la mysqldump is needed to
restore the database, restoring table definitions with ndb_restore will not work properly
    - Start fresh cluster
    - restore meta data via a mysql server
    - restore data via mysql server or ndb_restore as appropriate

Functionality added or changed:
    *ndb_restore will remove "trailing" event definitions
    *mgmapi stuctured events for ndbd exit
    *Overall new and improved error messaging on ndbd exit
    *DataMemory allocation moved first to ensure maximum available memory for malloc

Bugs fixed:
    *Event reports come out too often in the binlog
    *Stilled leak with growing Gci_container/mysqld (issues 6931,6990)
      + Bugfixes 5.0.14(2)-15 below


Alcatel Drop 5p4 (21 Sept 2005)

Functionality added or changed:
     Ndb files take the umask into account
     Mysqldump option --hex-blob will print bit-fields in hex format
     More trace info to be able to locate (if it should occur again) Bug #12594 Error
while reading the REDO log

Bugs fixed:
     (Bug #13179) partition info in mysqldump makes it hard to use dump in different sized
cluster
     (Bug #11367) Error message "Can't create table '." floods mysqld error log
      + Bugfixes 5.0.11-14(1) below


Alcatel Drop 5p3 (25 July 2005)

Functionality added or changed:
     Added more (and missing) error information.
     Enabled user defined partitioning for cluster, and a choice of modulo or linear
distribution
     Added load independant heartbeats

Bugs fixed:
     (Bug #11236) SQL thread incorrect handling of temporary errors for ndbcluster
     (Bug #11239) Partial master update of row that does not exist on slave breaks
replication
     (Bug #11300) epoch on slave is incorrect if -e option is used during metadata
restore.
     (Bug #11384) drop database causes mysqld to core
    *(Bug #11690) Weird error dropping database: REPL$BANK/ACCOUNT_TYPES Error Code: 4710
      + Bugfixes 5.0.10 below


Alcatel Drop 5p2 (9 June 2005)

Functionality added or changed:
     Support for distribution key together with binary collation
           Alcatel note: ensures that tables can be created with
                         distribution key whilst the ndb type being
                         char.

Bugs fixed:
     (Bug #11166) Inconsistency when applying backup log when inserts
     (Bug #10893) Cluster hangs on initial startup


Alcatel Drop 5p1 (8 June 2005)

Bugs fixed:
     *(Bug #11133) WriteTuple returns 626, if previous op in trans is a delete
                   Alcatel note: was the cause of the replication error sent on June 3.
     *(Bug #10190,updated to also handle distribution keys correctly) Backup from cluster
with NoOfReplica =1 is corrupt


Alcatel Drop 5 (3 June 2005)

Functionality added or changed:
      Replication supporting node recovery
      Frm format for partitioning changed (future proof)

      + Bugfixes 5.0.7 below


Alcatel Drop 4 (12 May 2005)

Functionality added or changed:
      Replication with limitation in node recovery support

      + Bugfixes and functionality addition up until 5.0.6 below + some from 5.0.7




Extract from the online web Change History with bugs relevant to Alcatels Cluster Usage
=======================================================================================

E.1.1. Changes in release 5.0.15 (Not yet released)

E.1.2. Changes in release 5.0.14(2) (Not yet released)
      (Bug #13512) NDB Cluster: If ndb_restore could not find a free mysqld process, it
crashed.
      (Bug #13053) NDB Cluster: Receipt of several enter single user mode commands by
multiple ndb_mgmd processes within a short period of time resulted in cluster shutdown.
      (Bug #12037) NDB Cluster: Multiple ndb_mgmd processes in a cluster would not know
each other's IP addresses.
      (Bug #11595) NDB Cluster: With two mgmd processes in a cluster, ndb_mgmd output for
SHOW would display the same IP address for both processes, even when they were on
different hosts.
      (Bug #13081) NDB Cluster: Queries on NDB tables that are executed using
index_merge/union or index_merge/intersection could produce incorrect results.
      (Bug #10694) NDB Cluster: LOAD DATA INFILE with a large data file failed.
      (Bug #13316) NDB Cluster: Adding an index to a table with a large number of tables
(more then 100) crashed the storage node.
E.1.2.1 Changes in release 5.0.14(1)
      (Bug #9282) API Node Crashes/Reloads on 'DELETE FROM'
     *(Bug #13152) BIT columns and following columns in NDB tables were corrupt when
dumped by mysqldump.

E.1.3. Changes in release 5.0.13 (Not yet released)
     ?(Bug #13054) start backup command return may be interrpted
     *(Bug #12992) Cluster StopOnError = Y restarts ndbd indefinitly
     *(Bug #12900) Missing reason for failed copy fragment
      (Bug #12608) Cluster shuts down during simultaneous recovery/commit
      (Bug #11607) error 4350 on deletion
      (Bug #10987) After NDBD failure, restart NDBD dies stating Unable to find restorable
replica
     ?(Bug #10950) Occational 'Error in mgm protocol parser' error on mysqld startup
      (Bug #10938, Bug #9996, Bug #11623) NDB: A cluster shutdown following the crash of a
data node would fail to terminate the remaining node processes, even though ndb_mgm showed
the shutdown request as having been completed.

E.1.4. Changes in release 5.0.12 (02 Sept 2005)

      (Bug #12833) NDB Cluster: Corrected the parsing of the CLUSTERLOG command by ndb_mgm
to allow multiple items.
      (Bug #11218) NDB Cluster: Improved error messages related to filesystem issues.
     *(Bug #12027) NDB Cluster: When a schema was detected to be corrupt, ndb neglected to
close it, resulting in a \x{201C}file already open\x{201D} error if the schema was opened
again later. written.
     *(Bug #12900) NDB Cluster: When it could not copy a fragment, ndbd exited without
printing a message about the condition to the error log. Now the message is written.
     *(Bug #12716) NDB Cluster: When a disk full condition occurred, ndbd exited without
printing a message about the condition to the error log. Now the message is written.
      (Bug #12118) NDB Cluster: An ALTER TABLE command caused loss of data stored prior to
the issuing of the command.
      (Bug #12220) NDB Cluster: Cluster failed to take character set data into account
when recomputing hashes (and thus could not locate records for updating or deletion)
following a configuration change and node restart.
      (Bug #10950) NDB Cluster: Wrong error message displayed when cluster management
server closed port while mysqld was connecting.

E.1.5. Changes in release 5.0.11 (06 Aug 2005)

      NDB: Improved handling of the configuration variables
NoOfPagesToDiskDuringRestartACC, NoOfPagesToDiskAfterRestartACC,
NoOfPagesToDiskDuringRestartTUP, and NoOfPagesToDiskAfterRestartTUP should result in
noticeably faster startup times for MySQL Cluster. (Bug #12149)


E.1.6. Changes in release 5.0.10 (27 July 2005)

Functionality added or changed:
      MySQL Cluster: A new -p option is available for use with the ndb_mgmd client. When
called with this option, ndb_mgmd prints all configuration data to stdout, then exits.

Bugs fixed:
      (Bug #11675) Invalid backup log when running with #nodes != 2^x
      (Bug #11355) server will not let me drop table with broken unique index
      (Bug #11942) Create/Drop table during backup causes cluster shutdown
      (Bug #11999) perror --ndb option documentation
      (Bug #12028) Hostname option in [NDB DEFAULT] section not used
      (Bug #11596) ndb_mgm doesn't show IP of connected nodes
      (Bug #11898) NDB_MGMD not releasing resources
      (Bug #11516) NDBCLuster: Server left core files following shutdown if data nodes had
failed.


E.1.7. Changes in release 5.0.9 (15 July 2005)

Bugs fixed:
      NDB Cluster: When trying to open a table that could not be discovered or unpacked,
cluster would return error codes which the MySQL server falsely interpreted as operating
system errors. (Bug #103651)


E.1.8. Changes in release 5.0.8 (not released)

Bugs fixed:
      MySQL Cluster: Connections between data nodes and management nodes were not being
closed following shutdown of ndb_mgmd. (Bug #11132)
      MySQL Cluster: mysqld processes would not reconnect to cluster following restart of
ndb_mgmd. (Bug #11221)
      MySQL Cluster: Fixed problem whereby data nodes would fail to restart on 64-bit
Solaris (Bug #9025)
      MySQL Cluster: Calling ndb_select_count() crashed the cluster when running on Red
Hat Enterprise 4/64-bit/Opteron. (Bug #10058)
      MySQL Cluster: Insert records were incorrectly applied by ndb_restore, thus making
restoration from backup inconsistent if the binlog contained inserts. (Bug #11166)
      MySQL Cluster: Cluster would time out and crash after first query on 64-bit Solaris
9. (Bug #8918)
      MySQL Cluster: ndb_mgm client show command displayed incorrect output after master
data node failure. (Bug #11050)
     *MySQL Cluster: A delete performed as part of a transaction caused an erroneous
result. (Bug #11133)
      MySQL Cluster: Not allowing sufficient parallelism in cluster configuration (e.g.
NoOfTransactions too small) caused ndb_restore to fail without providing any error
messages. (Bug #10294)
      MySQL Cluster: When using dynamically allocated ports on Linux, cluster would hang
on initial startup. (Bug #10893)
     *MySQL Cluster: Setting TransactionInactiveTimeout= 0 did not result in an infinite
timeout. (Bug #11290) 


E.1.9. MySQL Cluster-5.0.7 (10 june 2005)

Bugs fixed:
     *(Bug #11019) mgmapi start backup in some cases returns wrong backupid
      (Bug #9246) Condition pushdown and left join, wrong result
      (Bug #11050) ndb_mgm "show" prints incorrectly after master data node fails
      (Bug #10669) execute(NoCommit) freeze for ~6seconds
      (Bug #10956) More than 7 NR --initial, yields cluster failure
     *(Bug #10953) Read wo/ getValue during NF, could case system failure.
      (Bug #10948) Crash with ALTER/DROP table for NDB tables (HPUX and PPC32), duplicate
of Bug #9826
      (Bug #9826) Server crash on schema change ("drop table", "alter table") with NDB
      (Bug#9945) ALTER TABLE caused server crash. (Linux/390)
      (Bug#9826) (Bug #10948) Schema change (DROP TABLE, ALTER TABLE) crashed HPUX and
PPC32.
      (Bug#10711) (Bug#9363) (Bug#8918) (Bug#10058) (Bug#9025) Cluster would time out and
crash
                after first query; setting DataMemory to more than 2GB prevented cluster
from starting;
                calling ndb_select_count() crashed the cluster. (64-bit Unix OSes)


E.1.10. MySQL Cluster-5.0.6 (26 May 2005)

Functionality added or changed:
      Limit on number of metadata objects (number of tables, indexes and BLOBs) now
increased to 20,320 

Bugs fixed:
     *(Bug #10190) Backup from cluster wih NoOfReplica =1 is corrupt
                   Alcatel note: this is the issue that has been with ndb_restore
                                 complaining about distribution key
      (Bug #10193) Invalid DataDir in config causes ndbd segfault
      (Bug #10813) Build with SCI Transporter fails
      (Bug #10831) ndb mgmd LogDestination maxfiles does not rotate logs properly


E.1.11. MySQL Cluster-5.0.5 (not released)

Functionality added or changed:

      Decreased IndexMemory Usage
      Parallel key lookup (read-multi-range) for queries like SELECT * FROM t1 WHERE
primary_key IN (1,2,3,4,5,6,7,8,9,10);

Bugs fixed:
      (Bug #8315) NdbScanFilter cmp method only works for strings of exact word boundry
length
      (Bug #8103) Configuration handling error
      (Bug #8035) mysqld signal 10 when ndbd is shutdown
     *(Bug #7631) NDB$EVENT contains unreadable event and table names
     *(Bug #7628) Filtered event types are ignored
     *(Bug #7627) Drop Event operation fails
      (Bug #7424) create index on datetime fails
      (Bug #10471) Backup can become inconsisten wi/ certain combinations of multi updates
      (Bug #10287) ndb_select_all "delimiter" option non functional
      (Bug #10142) Unhandled resource shortage in UNIQUE index code
      (Bug #10029) crash in ordered index scan after db full
      (Bug #10001) 2 NDB nodes get signal 6 (abort) in DBTC
      (Bug #9969) 4012 - has misleading error message
     *(Bug #9960) START BACKUP reports failure albeit succeding
      (Bug #9924) ABORT BACKUP 1 crashes 4 node cluster
      (Bug #9892) Index activation file during node recovery
      (Bug #9891) Crash in DBACC (line 7004) during commit
      (Bug #9865) SELECT does not function properly
      (Bug #9839) Column with AUTOINC contains -1 Value on node stop
      (Bug #9757) Uncompleted node failure after gracefully stopping node
      (Bug #9749) Transactions causes deadlock in ACC
      (Bug #9724) Node fails to start: Message: File has already been opened
      (Bug #9691) UPDATE fails on attempt to update primary key
      (Bug #9675) Auto-increment not working with INSERT..SELECT and NDB storage
      (Bug #9318) drop database does not drop ndb tables
      (Bug #9280) Memory leak in cluster when dependent sub-queries are used
      (Bug #8928) create table with keys will shutdown the cluster
      (Bug #8928) Creating a table did not work for a cluster with 6 nodes. Databases with
1, 2, 4, 8, ... (2^n nodes) did not have the problem. After a rolling upgrade, restart
each node manually by restarting it with the --initial option. Otherwise, use dump and
restore after an upgrade.
      (Bug #9916) DbaccMain.cpp / DBACC (Line: 4876) / Pointer too large
      (Bug #9435) TIMESTAMP columns don't update
      (Bug #9052) Uninitialized data during unique index build, potential cluster crash
      (Bug #8876) Timeout when committing aborted transaction after node failure
      (Bug #8786) ndb_autodiscover, drop index can fail, wait 2 minutes timeout
      (Bug #8853) Transaction aborted after long time during node failure (4012)
      (Bug #8753) Invalid schema object version after dropping index (crash fixed,
currently retry required)
      (Bug #8645) Assertion failure with multiple management servers
      (Bug #8557) ndbd does not get same nodeid on restart
      (Bug #8556) corrupt ndb_mgm show printout for certain configurations
      (Bug #8167) cluster shared memory and mysqld signal usage clash


E.1.12. MySQL Cluster-5.0.4 (16 Apr 2005)

Functionality added or changed:
      Condition pushdown to storage engine now works for update and delete as well 

Bugs fixed:
      (Bug #9675) Auto-increment not working with INSERT..SELECT and NDB storage
      (Bug #9517) Condition pushdown to storage engine does not work for update/delete
      (Bug #9282) API Node Crashes/Reloads on 'DELETE FROM'
      (Bug #9280) Memory leak in cluster when dependent sub-queries are used
      (Bug #8585) ndb_cache2 fails on aix52


E.1.13. MySQL Cluster-5.0.3 (23 Mar 2005: Beta)

Functionality added or changed:
      Condition pushdown to storage engine
      Query cache enabled for cluster
Bugs fixed:
      (Bug #8284) Out of fragment memory in DBACC
      (Bug #8262) Node crash due to bug in DBLQH
      (Bug #8208) node restart fails on Aix 5.2
      (Bug #8167) cluster shared memory and mysqld signal usage clash
      (Bug #8101) unique index and error 4209 while selecting
      (Bug #8070) (Bug #7937) (Bug #6716) various ndb_restore core dumps on HP-UX
      (Bug #8010) 4006 forces MySQL Node Restart
      (Bug #7928) out of connection objects
      (Bug #7898) mysqld crash with ndb (solaris)
      (Bug #7864) Not possible to have more than 4.5G data memory