Version End of Life. Not Yet Set
Release 7.0.2 contains a number of key bug fixes and improvements.
The following changes have been made to Tungsten Cluster and may affect existing scripts and integration tools. Any scripts or environment which make use of these tools should check and update for the new configuration:
repl_svc_extractor_multi_frag_service_detectionis now turned ON by default. Event shards are determined at extraction time. With fragmented events, the shard cannot be determined by only reading the first fragment, but needs to check the last fragment as well. With this setting turned OFF, there is no issue with pipelines that don't need it, i.e. no parallel apply downstream replicas. However, as this is done at extract time, THL contains this information, and adding or changing a replica using parallel apply could introduce issues.
It can be disabled if you see a performance overhead but this should be done with caution. For Aurora<>Aurora Active/Active deployments it is essential that this property be left ON.
No longer sending write traffic to the remote site unless the local site is fully offline. In case of local failover, the connector will now pause connections until a new primary is elected. This will avoid risks of out-of-order apply after local failover
rsync is now an option in tprovision, in addition to using xtrabackup and mysqldump. To use rsync, specify
Using rsync by default will provision a replica in 2 passes:
The first pass will live copy (seed) the replica from the source.
The second pass will quiesce the source and run the rsync again, resulting in shorter down time than a single pass rsync
-c option is now available with some trepctl commands that can be used in conjunction
-r option to indicate the number of times to refresh before automatically terminating.
For example, the following command:
trepctl perf -r 3 -c 10
Will refresh the output every 3 seconds, 10 times.
The trepctl status will now show the last known applied seqno and latency.
This information is stored on disk at regular intervals (10s minimum) so as not to overload the replicator, therefore the value can be shown as slightly old dependant on when the status command was issued.
By default, this feature is disabled. It can be enabled by setting the following parameter in the configuration :
This will write current position to disk every 20 seconds. This information is also exported by the Prometheus exporter.
If the service is online, it will display the current value (the same as appliedLastSeqno and appliedLatency)
Processing status command...
appliedLastEventId : mysql-bin.000017:0000000151329854;70
appliedLastSeqno : 999
appliedLatency : 347707.0
lastKnownAppliedSeqno : 999
The tungsten_merge_logs command now supports the
--before TIMESTAMP and
--after TIMESTAMP filters
an update statement was logged on primary, but did not update any row on replica
a delete statement was logged on primary, but did not delete any row on replica
Also, tpm ask now supports direct calls to coordinator,
The tungsten_generate_haproxy_for_api and tpm generate-haproxy-for-api commands
now support using connector hosts in the backend definitions via
-c, and extra backend flags to the
backend hosts lines using
The tungsten_generate_haproxy_for_api and tpm generate-haproxy-for-api
commands no longer call the
Perl Data::Dumper module.
The tungsten_reset_manager command now supports the ability to simply print out the path or paths to be cleared,
one per line via the
The tmonitor command now accepts cli args to specify the ports and will auto-configure the ports if they have been changed via the Tungsten configuration.
The tpm command calls to
glob have been improved to be more strict and compliant.
The tpm ask command has five new variables available:
dsstate for the current datasource,
trstate for the current replicator, and
nodeinfo which displays all 4 of the new variables.
A new standalone status script has been added called tungsten_get_status that shows the datasources and replicators for all nodes in all services along with seqno and latency.
The tpm report command now prints the hostname and listener ports where available when using the
--extra|-x option or the new
Full details and examples of these new options can be found here
This will pause the given stage for 100 seconds.:
trepctl pause -stage thl-to-q -time 100
This will pause the stage indefinitely (or until restart, etc) Add -y to avoid the prompt message whether you are sure.
trepctl pause -stage thl-to-q
For the previous 2 commands, running a pause command again will override the previous command.
This will resume the suspended stage (Note that if the stage is not paused, this will have no effect):
trepctl resume -stage thl-to-q
Please note this pause does not survive a replicator restart or a service offline/online.
Per-service tuning of the replicator thl directory is now possible for multi-service replicator-only installs as well as for clustering. The given value should be the base directory, to which tungsten will add the service name. For example, the following entry in the tungsten.ini:
[alpha] ... ... thl-directory=/drv1/thl ...
Would result in the THL being placed in /drv1/thl/alpha
Update of thl directory is only available when tpm is called from the staging installation directory, NOT from the running directory.
For these properties to be in effect, you must ensure that
optimize-row-events=true is either explicitly set
in your configuration, or not present (since it will be enabled by default)
For example, the following settings will limit the number of inserted or deleted rows applied at once to 10:
The default values if not specified will be 50 for inserts and 100 for deletes. Note that for deletes to be optimized, the affected table MUST have a single column PK.
The connector graceful-stop command now supports systemd service manager properly. The connector stop command now takes an optional argument that will make it a graceful stop. If connector stop is run without the parameter, it will stop the connector immediately. If a positive number of seconds is passed, it will wait, at most, this timeout for connections to disconnect (refusing new connections), after which it will force close all connections and shutdown the connector. connector graceful-stop behavior is unchanged: without the parameter, the connector will wait "forever" for connections to disconnect. A positive timeout in seconds can be passed to sever connections after the given delay
Two new commands have ben introduced to cctrl to provide better insight and control of the connectors. These are:
datasource <dsname> connections [-l]
This command displays the current number of connections running to the given node through connectors.
datasource <dsname> drain [optional timeout]
This command will prevent new connection to be made to the given data source, while ongoing connection remain untouched.
For more detils on the usage and options, see cctrl datasource Command
Added a new option to TPM
manager-replicator-offline-timeout=<timeout_in_sec> that configures the timeout for
the manager to wait until the replicator goes offline. When parallel applier is in use the default timeout was too low, so it’s now
user configurable so that it can be adjusted to suit different topologies. If not supplied, the default is 180 (3 minutes). This
value should be sufficent in most use cases.
The tungsten_skip_seqno command no longer fails when
-i is specified, and now properly filters using
--filter when there is a long error message.
The tpm command now allows any case for section entries (i.e. [alpha_FROM_beta]) in the INI files.
The tpm diag command now gathers the
mysql.log file when SSL is enabled in the server.
The tpm command checks for the existence of the mysql command-line client when installing/upgrading. The process will no longer abort with an error on non-MySQL targets such as heterogeneous replicator appliers, or Active-Witness hosts.
Issues: CT-1924, CT-2018
Fixes an issue that prevented dsctl from connecting to MySQL if SSL was enabled.
The tpm mysql command will now gracefully handle being run on a non-database node.
The tpm mysql command no longer aborts with an access denied error on CentOS 6.
The cluster_backup script will no longer backup a replica if the replicator is in an ERROR state.
Fixes an issue that would prevent a service from going offline at a specified time (trepctl online -until-time) when parallel apply is enabled. This is a rework of CT-1243.
Fixed a possible issue when recovering an old primary as a replica after failover when parallel apply is enabled, that could lead the replica to be unable to come online and require a reprovisioning of this replica.
Fixed an issue where filtered events would trigger a useless update to the service trep_commit_seqno table while it is overwritten anyway once the last statement of the applied event is done, just prior to committing the whole block.
A bug has been fixed that, in a few very rare cases, would allow replicas to continue to pull and apply THL from a failed primary whilst a failover was in the process of electing a new primary. This resulted in failovers being unable to complete fully. Whilst the new primary would be online and functioning, existing replicas in the cluster could experience errors due to THL discrepencies between the old and new primary nodes.