codership / documentation Goto Github PK

The code contains this option but I could not find any documentation about it anywhere.

Overwiew:

Auto eviction is a process to remove a node with erratic behavior permanently from cluster.

Each of the nodes monitor group communication response times from other nodes and if one of the other nodes gets delayed for more than evs.delay_margin from expected response time, it will be added to the list of delayed nodes. If node on delayed list stays responsive for evs.delayed_keep_period time, one entry is removed from delayed list. If the number of entries in delayed list for some node reaches evs.auto_evict and majority of the nodes have it on delayed list, the node is evicted permanetly from the cluster and cannot join back until restarted.

Related Parameters:

evs.delay_margin:

Time period that a node can delay its response from expected until it is added to delayed list. The value must be higher than the highest RTT between nodes. Default PT1S.

evs.delayed_keep_period:

Time period that node is required to remain responsive until one entry is removed from delayed list. Default PT30S.

evs.evict

Manual eviction can be triggered by setting evs.evict to certain node value. Setting evs.evict to empty string will clear the evict list on the node where it is set.

evs.auto_evict:

Number of entries allowed on delayed list until auto eviction takes place. Setting value to 0 disables auto eviction protocol on the node, though node response times will still be monitored. Default 0.

evs.version:

EVS protoco version. Auto eviction is enabled on evs.version=1. Default 0 for backwards compatibility.

New Status Variables:

The following status variables have been added:

wsrep_evs_state:

Internal EVS protocol state.

wsrep_evs_delayed:

Comma separated list of nodes that are considered delayed. The node format is <uuid>:<address>:<count> where <count> is the number of entries on delayed list for that node.

wsrep_evs_evict_list:

List of UUIDs of the evicted nodes.

Upgrade From Previous Versions:

In order to take advantage of auto eviction, EVS protocol must be upgraded to version 1.

Versions before 3.8 used evs.version=0 and are not directly compatible with evs.version=1. Upgrade from previous versions must be in the following way:

• Start upgraded node with evs.version=0 in wsrep_provider_options
• Once all nodes have been upgraded, set evs.version=1 in
  wsrep_provider_options on all of the nodes.
• Add evs.version=1 to wsrep_provider_options in configuration to make
  sure that version upgrade is maintained in the following restarts.

Section http://galeracluster.com/documentation-webpages/galerainstallation.html#disabling-selinux-for-mysqld links to https://blogs.oracle.com/jsmyth/selinux_and_mysql which returns 404.

Issue by johankooi
Thursday Jun 05, 2014 at 10:02 GMT
Originally opened as codership/galera#52

On Fedora (and maybe some more distributions) there is in the /etc/sudoers file an line which says:
Defaults requiretty
You must comment out this line before you can start garbd because in the /etc/init.d/garb there is a sudo -u nobody statement which fails on the standard sudoers file of Fedora.

When referencing the database configuration documentation [1], there is a typo in the following line:

wsrep_provider_options="gcache.size=32G; gachce.page_size=1G"

which I believe should be:

wsrep_provider_options="gcache.size=32G; gcache.page_size=1G"

[1] https://github.com/codership/documentation/blob/master/galeracluster/source/dbconfiguration.rst#basic-wsrep-provider-settings

$ make dirhtml

produces the following at the final stage:

galeracluster/source/createcluster.rst:125: WARNING: undefined label: weighted quorum (if the link has no caption the label must precede a section header)
galeracluster/source/dealingwithmultimasterconflicts.rst:10: WARNING: undefined label: certification based replication (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:55: WARNING: undefined label: galera arbitrator (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:12: WARNING: undefined label: galera replication plugin (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:25: WARNING: undefined label: wsrep api (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:45: WARNING: undefined label: incremental state transfer (ist) (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:30: WARNING: undefined label: primary component (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:60: WARNING: undefined label: rolling schema upgrade (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:40: WARNING: undefined label: state snapshot transfer (sst) (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:65: WARNING: undefined label: total order isolation (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:50: WARNING: undefined label: writeset cache (gcache) (if the link has no caption the label must precede a section header)
galeracluster/source/glossary.rst:35: WARNING: undefined label: wsrep api (if the link has no caption the label must precede a section header)
galeracluster/source/index.rst:18: WARNING: term not in glossary: galera replicator
galeracluster/source/nodeprovisioning.rst:34: WARNING: term not in glossary: seqno
galeracluster/source/nodeprovisioning.rst:34: WARNING: term not in glossary: incremental state transfer
galeracluster/source/statetransfer.rst:49: WARNING: undefined label: comparison-of-state-transfer-methods (if the link has no caption the label must precede a section header)
galeracluster/source/weightedquorum.rst:10: WARNING: undefined label: galera arbitrator (if the link has no caption the label must precede a section header)

I supposed it's a bug.

$ make dirhtml

produces the following warnings:

galeracluster/source/compiling.rst:: WARNING: document isn't included in any toctree
galeracluster/source/createcluster.rst:: WARNING: document isn't included in any toctree
galeracluster/source/installmariadbdeb.rst:: WARNING: document isn't included in any toctree
galeracluster/source/installmariadbrpm.rst:: WARNING: document isn't included in any toctree
galeracluster/source/installmysqldeb.rst:: WARNING: document isn't included in any toctree
galeracluster/source/installmysqlrpm.rst:: WARNING: document isn't included in any toctree
galeracluster/source/installxtradbdeb.rst:: WARNING: document isn't included in any toctree
galeracluster/source/installxtradbrpm.rst:: WARNING: document isn't included in any toctree
galeracluster/source/sysconfiguration.rst:: WARNING: document isn't included in any toctree

Quick examination shows that some of those files contain redundant information. So the question is: are they outdated and were consciously dropped from TOC? Or is it the opposite and they are new and should supersede other files?

If they are outdated, why they are still in the tree? If they are needed for reference, they probably should be excluded from compilation then.

If they are new, then we have a bug, since they are not visible/accessible through HTML links.

how to generate a policy

semanage port -a -t mysqld_port_t -p tcp 4567
semanage port -a -t mysqld_port_t -p udp 4567
semanage port -a -t mysqld_port_t -p tcp 4444

semanage permissive -a mysqld_t
restart mysqld

# do stuff to generate a IST a SST and anything else (notify script)?.

fgrep mysqld /var/log/audit/audit.log | audit2allow  -m MySQL_galera -o MySQL_galera.te
more MySQL_galera.te

# edit as required.

checkmodule -M -m MySQL_galera.te -o  MySQL_galera.mod
semodule_package -m MySQL_galera.mod -o MySQL_galera.pp   
semodule -i MySQL_galera.pp

# remove permissive mode:
semanage permissive -d mysqld_t

Neither bugs, code or downloads will be hosted on launchpad any longer.

We have a wiki , http://wiki.codership.com/, which may contain some information not available in the documentation, such as a description of the SST protocol and the conventions for writing new SST scripts.

All content should be integrated into the documentation so that the wiki.codership.com host can be retired.

Migrate galera/docs directory and document labeled issues from codership/galera repo. Docs will be stored under galeracluster directory.

Issue by kennethpjdyer
Thursday May 08, 2014 at 02:18 GMT
Originally opened as codership/galera#18

Change the Galera documentation over to the new theme.

• Adjust the CSS for better readability.
• Push the changes out.
• Fix and add the Edit on Github link to the lower navigation bar.

Issue by grypyrg
Tuesday Jun 17, 2014 at 10:11 GMT
Originally opened as codership/galera#62

wsrep_desync is missing from http://galeracluster.com/documentation-webpages/mysqlwsrepoptions.html

Issue by temeo
Tuesday May 06, 2014 at 14:31 GMT
Originally opened as codership/galera#13

The page says:

"Galera Cluster backups can be performed just as regular MySQL backups, using a backup script. Since all the cluster nodes are identical, backing up one node backs up the entire cluster.

However, such backups will have no global transaction IDs associated with them. You can use these backups to recover data, but they cannot be used to recover a Galera Cluster node to a well-defined state. Furthermore, the backup procedure may block the cluster operation for the duration of backup, in the case of a blocking backup."

However

• Blocking cluster operation can be avoided with wsrep_desync option which is present in resent versions of Galera Cluster
• In case of rsync backup Global Transaction ID can be recovered from backed up InnoDB tablespace
• Xtrabackup should have option to record GTID consistently

This section should be expanded to cover ways to take backups without the aid of garbd.

Rendering the documentation runs into the following errors, among others, which cause parts of it to go missing:

/home/philips/git/documentation/galeracluster/source/index.rst:71: ERROR: Error in "toctree" directive:
unknown option: "includehidden".

.. toctree::
   :includehidden:
   :maxdepth: 3

   technicaldescription
   gettingstarted
   userguide
   troubleshooting
   reference

I am not sure why is that and if it is a bug in sphinx. Removing the line causes the problem to go away.

/home/philips/git/documentation/galeracluster/source/monitoringthecluster.rst:67: ERROR: Error in "code-block" directive:
1 argument(s) required, 0 supplied.

.. code-block::

     SHOW VARIABLES LIKE 'wsrep_cluster_status';

This is a genuine bug in the documentation source.

checking consistency... /home/philips/git/documentation/galeracluster/source/compiling.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/createcluster.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/installmariadbdeb.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/installmariadbrpm.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/installmysqldeb.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/installmysqlrpm.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/installxtradbdeb.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/installxtradbrpm.rst:: WARNING: document isn't included in any toctree
/home/philips/git/documentation/galeracluster/source/sysconfiguration.rst:: WARNING: document isn't included in any toctree

So those documents will not be linked from the TOC?

Issue by grypyrg
Tuesday Jul 08, 2014 at 06:44 GMT
Originally opened as codership/galera#77

=== https://bugs.launchpad.net/percona-xtradb-cluster/+bug/1338523

DOC: http://galeracluster.com/documentation-webpages/galerastatusvariables.html#wsrep-flow-control-sent

in 25.3.1 galera behavior changed for these counters: https://bugs.launchpad.net/galera/+bug/1180792
They no longer reset every time the query runs.

and limitations thereof.

http://galeracluster.com/documentation-webpages/configurationtips.html#wan-replication

Parameter evs.inactive_check_period = PT10S; should not be changed from the default, so it should removed from the wsrep_provider_options.

Configuration option

wsrep_provider_options="evs.join_retrans_period=14"

has a wrong value format. According to reference it uses ISO8601 duration format. Better value would be:

wsrep_provider_options="evs.join_retrans_period=PT0.5S"

Issue by ronin13
Friday Jul 25, 2014 at 10:56 GMT
Originally opened as codership/galera#84

Purpose and (internal) use of this file is not documented.

Issue by kennethpjdyer
Tuesday May 20, 2014 at 22:12 GMT
Originally opened as codership/galera#36

The Sphinx build system by default writes in numbering on section headers. Trim this down for a cleaner look.

Adapt the header pages to the new format. Replace links with :doc: elements. Use a hidden toctree to keep the navigation links in line.

Issue by ronin13
Thursday Jul 03, 2014 at 15:42 GMT
Originally opened as codership/galera#75

The automatic donor selection in 3.6 based on gcache seqno limits and IST/SST accordingly is not documented.

$ make dirhtml

produces in the log:
galeracluster/source/managingfc.rst:113: WARNING: Duplicate explicit target name: "managing-fc".
There are really two such targets in the file and I guess it is a matter of hyperlinks and TOC: one of them should be either renamed or removed.

We now have our own package repositories, so the download installation instructions need to be updated to reflect that fact.

For CentOS, Fedora:

1. Create a file /etc/yum.repos.d/galera.repo with the following contents:

[galera]
name=Galera
baseurl=http://releases.galeracluster.com/DIST/RELEASE/x86_64
gpgcheck=1
gpgkey=http://releases.galeracluster.com/GPG-KEY-galeracluster.com

replacing DIST with centos or fedora and RELEASE is the OS release number, e.g. 6,7 for Centos , 20, 21 for Fedora

1. Install packages:

sudo yum install galera-3
sudo yum install mysql-wsrep-5.6

Package galera-3 will install Galera wsrep provider and garbd. Package mysql-wsrep-5.6 will install MySQL server and client components.

For OpenSuse, SLES:

1. Import GPG key:

sudo rpm --import "http://releases.galeracluster.com/GPG-KEY-galeracluster.com"

1. Create a galera.repo file in local directory.

[galera]
name=Galera
baseurl=http://releases.galeracluster.com/DIST/RELEASE/x86_64
gpgcheck=1

replacing DIST with opensuse, or sles and RELEASE with version number

1. Install repository:

    sudo zypper addrepo galera.repo
    sudo zypper refresh

    sudo zypper install galera-3
    sudo zypper install mysql-wsrep-5.6

For Ubuntu, Debian:

1. Install key

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv BC19DDBA

1. Install the Galera package repository.

Create a file named /etc/apt/sources.list.d/galera.list with the following content:

deb http://releases.galeracluster.com/DIST RELEASE main

where DIST is either ubuntu or debian and RELEASE is the release codename (precise, trusty for Ubuntu, wheezy for Debian).

1. Install packages:

    sudo apt-get update --ignore-missing
    sudo apt-get install galera-3 galera-arbitrator-3
    sudo apt-get install mysql-wsrep-5.6

Google does not return any documentation results from galeracluster.com, regardless of the search keyword.

I submitted a request to google to reindex the domain, but I think there may be a systematic problem that is keeping those pages out of the Google index.

Relates to firewallsettings.rst and galerainstallation.rst.

Revise the firewall configuration sections in the docs. Reorganize the information displayed to allow for expansion to include one or two other firewall tools as an alternative to iptables, (e.g. nftables if it's ready and pf).

Update the iptables documentation:

• Organize LAN and WAN examples into subsections.
• Completely revise the text and commands for saving tables, (right now they're wrong).
• Update all code-blocks to use the more verbose style arguments, (.e.g., --append vs. -A).
• Improve text to explain what each part of the command does, given that the reader is expected to work out a better format to suit their own systems.

The MySQL portion of the OpenStack manual has a section " Configure OpenStack services for highly available MySQL"

http://avoceteditors.com/private/codership/openstack/high-availability-guide/content/_configure_openstack_services_for_highly_available_mysql.html

which contains information on what setting to use in the place of a single-node MySQL DSN when HA is used. I think we should have a similar section for Galera, describing the correct format for the DSN that would provide proper redundancy.

under galera tree:

https://github.com/codership/galera/blob/3.x/docs/source/index.rst

For each installation page in Getting Started, there is an entry towards the end that covers system table upgrades. This to be done in the event that the user install Galera Cluster over an existing standalone instance of MySQL.

Remove this entry from each page. Create a new entry in the User Guide, add seealso block to refer people to it from Getting Started. mysql_upgrade should be run when the node is not part of the cluster. Start the node with no wsrep Provider, (from command line with argument wsrep_on=OFF?). Run mysql_upgrade. Stop the node. Start the node with wsrep provider and cluster address, allowing it to join the cluster.

Issue by kennethpjdyer
Monday Jun 16, 2014 at 01:15 GMT
Originally opened as codership/galera#61

Update the Administrative Guide for new text and formats.

Issue by temeo
Tuesday May 06, 2014 at 12:19 GMT
Originally opened as codership/galera#12

Originally reported here: https://groups.google.com/forum/#!topic/codership-team/O4vHuGBKRC0

http://galeracluster.com/documentation-webpages/galeraparameters.html - 'replication.*' should be 'repl.*'.

missing options on this page:

• socket.checksum
• repl.key_format
• repl.proto_max
• repl.max_ws_size
• pc.announce_timeout
• pc.wait_prim_timeout
• gmcast.segment
• base_host
• base_port
• cert.log_conflicts

other suggestions:

4.16.6 selinux - a link to https://blogs.oracle.com/jsmyth/entry/selinux_and_mysql would probably help

5.3.1. Diagnosing Multi-Master Conflicts

could include a reference to cert.log_conflicts

https://github.com/codership/documentation/blob/master/galeracluster/source/galerastatusvariables.rst

Issue by temeo
Tuesday Jun 17, 2014 at 13:29 GMT
Originally opened as codership/galera#65

3.x branch has a new wsrep provider option pc.recovery (boolean). If the value for this parameter is set to true, primary component state is stored to disk and recovered from there when node is started.

Primary component is recovered automatically when all of the nodes that were present in the last known primary component have established communication between each other.

This feature solves two use cases:

• Automatic recovery from full cluster crash (for example full data center power outage)
• Graceful full cluster restart without the need for bootstrapping new primary component explicitly

Known limitations:

• If the wsrep position differs between nodes during the recovery, full SST is required
• ...

Issue by zba
Monday Jul 28, 2014 at 04:14 GMT
Originally opened as codership/galera#87

I tryed with centos6 , followed this

Centos6 has no mysql-client package, also if I try to install mysql-community-client from oracle repository it tells about conflicts, do you provide actual documentation for rhel6/centos6 ?

also what means strings like this ?

3. Install the compatibility libraries:

rpm -e mysql-shared-compat.rpm

doesn't it should be like

  rpm -i mysql-shared-compat-*.rpm 

or something like this ?

Issue by dirtysalt
Tuesday Aug 12, 2014 at 07:33 GMT
Originally opened as codership/galera#95

This variable is introduced in patch https://bugs.launchpad.net/codership-mysql/+bug/1277053

Originally, causality check could only be applied on some SQL statement like SELECT, BEGIN/END, SHOW STATUS etc, but not on some autocommit UPDATE/DELETE/REPLACE/INSERT SQL statement.

This variable is introduced to control causality check on those statements. It's designed as bit mask.

"wsrep_sync_wait", "Ensure "synchronous" read view before executing an operation of the type specified by bitmask: 1 - READ(includes SELECT, SHOW and BEGIN/START TRANSACTION); 2 - UPDATE and DELETE; 4 - INSERT and REPLACE"

Variable 'wsrep_causal_reads' also is kept for backward compatibility, but is deprecated. User could turn on/off READ flag on 'wsrep_sync_wait' which is equivalent to turn on/off 'wsrep_causal_reads'.

The Flow Control page under Technical Description partly describes how it works and partly documents its management. Break this up, to retain the description while moving the management chapter into the user guide.

Issue by temeo
Tuesday Sep 09, 2014 at 07:50 GMT
Originally opened as codership/galera#114

As discussed here:
https://bugs.launchpad.net/percona-xtradb-cluster/+bug/1359700

gcache.mem_size is not supported and causes crashes. Either it should be removed from the allowed configuration options and documentation or the crashes should be fixed.

http://galeracluster.com/documentation-webpages/arbitrator.html#starting-galera-arbitrator-from-the-shell

$ garbd --group=example_cluster \
     --address="192.168.1.1 192.168.1.2 192.168.1.3" \
     --option="socket.ssl_cert=/etc/galera/cert/cert.pem"

should be

$ garbd --group=example_cluster \
     --address="gcomm://192.168.1.1,192.168.1.2,192.168.1.3" \
     --option="socket.ssl_cert=/etc/galera/cert/cert.pem"

and

# arbtirator.config
group = example_cluster
address = "192.168.1.1 192.168.1.2 192.168.1.3"

should be

# arbtirator.config
group = example_cluster
address = "gcomm://192.168.1.1,192.168.1.2,192.168.1.3"

In the Getting Started pages, the user configures their first cluster to use mysqldump. Update these entries so that they start out using rsync.

Rework and expand the Admin Guide entries on SST. They need better organization, cleaner text and specific information on setting up mysqldump for use.

Issue by grypyrg
Tuesday Jun 03, 2014 at 07:22 GMT
Originally opened as codership/galera#48

wsrep_flow_control_paused_ns is not documented on http://galeracluster.com/documentation-webpages/galerastatusvariables.html#wsrep-flow-control-paused

PXC 5.6 has a brief mention about it: http://www.percona.com/doc/percona-xtradb-cluster/5.6/wsrep-status-index.html

I can't find wsrep_load_data_splitting on http://galeracluster.com/documentation-webpages/mysqlwsrepoptions.html

This page

http://galeracluster.com/documentation-webpages/arbitrator.html

I think mixes the two concepts:

• starting garbd from the command line, passing it options on the command line
• starting garbd as a service, where the options are placed in a configuration file.

The location of the configuration file is not mentioned (it can be different by distro), and the example of a configuration file that is given is wrongly called "init script".

Issue by kennethpjdyer
Friday Apr 25, 2014 at 21:45 GMT
Originally opened as codership/galera#4

At the moment, symptomsolution.rst renders as a table where the particular symptom is listed in the left column and a very detailed solution is listed in the right column. While the logic of this is pretty clear, on a webpage the table formatting makes it difficult to read and find what you're looking for. Additionally, the table entries do not posses id references, so, in the event of a client having difficulties, it's not possible to link them to a particular symptom / solution pairing on the page.

Rewrite the table material into separate sections. If there are enough entries to justify it, set up a link table like on the parameters pages.

Update Reference chapter pages. Improve text, clean up tables, add examples, et cetera.

Issue by philip-galera
Monday Sep 15, 2014 at 07:40 GMT
Originally opened as codership/galera#125

Searching using the documentation search box never returns any results. A possible workaround would be to redirect the documentation search box to use the general site search, which does seem to return results.

Review and update the Technical Description chapter.

Issue by kennethpjdyer
Saturday Apr 26, 2014 at 02:24 GMT
Originally opened as codership/galera#6

The old configuration page carried a section on failure simulation that made three general points on ways in which it could be done. Expand this into a full tutorial on how to break the cluster in controlled ways to test replication and prepare for actual breakage. Add the new page to the Getting Started Guide, either as a section on testingcluster.rst or a separate page just after.

Issue by temeo
Thursday Jun 19, 2014 at 13:25 GMT
Originally opened as codership/galera#67

Docoment new status variable wsrep_evs_repl_latency. The variable tells group communication replication latency. The latency is measured from the time point when message is sent to the time point when message is delivered.

The status variable format is the following:

wsrep_evs_repl_latency       | 0.00243433/0.144022/0.591963/0.215824/13

Fields are min/average/max/standard deviation/sample size.