Clustering

v1.3.9 [2018-01-19]

Bugfixes

• Changed handling of previous stream closed error within io.EOF. Now dealt with as remote iterator interrupt.
• Fix spurious rpc error: i/o deadline exceeded errors.
• Discard remote iterators that label their type as unknown.
• Do not queue partial write errors to hinted handoff.

v1.3.8 [2017-12-04]

Upgrading – for users of the TSI preview

If you have been using the TSI preview with 1.3.6 or earlier 1.3.x releases, you will need to follow the upgrade steps to continue using the TSI preview. Unfortunately, these steps cannot be executed while the cluster is operating – so it will result in downtime.

See more details here: Upgrading from previous versions

Bugfixes

• Updated yamux resolves race condition when SYN is successfully sent and a write timeout occurs.
• Resend authentication credentials on redirect
• Fix wildcard when one shard has no data for a measurement with partial replication.
• Fix spurious rpc error: i/o deadline exceeded errors.

v1.3.7 [2017-10-26]

Upgrading – for users of the TSI preview

The 1.3.7 release resolves a defect that created duplicate tag values in TSI indexes. See Issues #8995, and #8998. However, upgrading to 1.3.7 cause compactions to fail, see Issue #9025. We will provide a utility that will allow TSI indexes to be rebuilt, resolving the corruption possible in releases prior to 1.3.7. If you are using the TSI preview, you should not upgrade to 1.3.7 until this utility is available. We will update this release note with operational steps once the utility is available.

Bugfixes

• Read for the interrupt signal from the stream before creating the iterators.
• Address Deadlock issue in meta server on 1.3.6
• Fix logger panic associated with anti-entropy service and manually removed shards.

v1.3.6 [2017-09-28]

Bugfixes

• Fix “group by” marshaling in the IteratorOptions.
• Address meta service data race condition.
• Fix race condition when writing points to remote nodes.
• Use InfluxData fork of yamux. This update reduces overall memory usage when streaming large amounts of data. Contributed back to the yamux project via: https://github.com/hashicorp/yamux/pull/50
• Address data race reading Len() in connection pool.

v1.3.5 [2017-08-29]

This release builds off of the 1.3.5 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS releases.

v1.3.4 [2017-08-23]

This release builds off of the 1.3.4 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS releases.

Bugfixes

• Close connection for remote iterators after EOF to avoid writer hanging indefinitely

v1.3.3 [2017-08-10]

This release builds off of the 1.3.3 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS releases.

Bugfixes

• Connections are not closed when CreateRemoteIterator RPC returns no iterators, resolved memory leak

v1.3.2 [2017-08-04]

Bugfixes

• influxd-ctl restore -newdb unable to restore data.
• Improve performance of SHOW TAG VALUES.

Bugfixes

• Show a subset of config settings in SHOW DIAGNOSTICS.
• Switch back to using cluster-tracing config option to enable meta HTTP request logging.
• Fix remove-data error.

v1.3.1 [2017-07-20]

Bugfixes

• Show a subset of config settings in SHOW DIAGNOSTICS.
• Switch back to using cluster-tracing config option to enable meta HTTP request logging.
• Fix remove-data error.

v1.3.0 [2017-06-21]

Configuration Changes

[cluster] Section

• max-remote-write-connections is deprecated and can be removed.
• NEW: pool-max-idle-streams and pool-max-idle-time configure the RPC connection pool. See config.sample.toml for descriptions of these new options.

Removals

The admin UI is removed and unusable in this release. The [admin] configuration section will be ignored.

Features

• Allow non-admin users to execute SHOW DATABASES
• Add default config path search for influxd-meta.
• Reduce cost of admin user check for clusters with large numbers of users.
• Store HH segments by node and shard
• Remove references to the admin console.
• Refactor RPC connection pool to multiplex multiple streams over single connection.
• Report RPC connection pool statistics.

Bugfixes

• Fix security escalation bug in subscription management.
• Certain permissions should not be allowed at the database context.
• Make the time in influxd-ctl’s copy-shard-status argument human readable.
• Fix influxd-ctl remove-data -force.
• Ensure replaced data node correctly joins meta cluster.
• Delay metadata restriction on restore.
• Writing points outside of retention policy does not return error
• Decrement internal database’s replication factor when a node is removed.

v1.2.5 [2017-05-16]

This release builds off of the 1.2.4 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS releases.

Bugfixes

• Fix issue where the ALTER RETENTION POLICY query does not update the default retention policy.
• Hinted-handoff: remote write errors containing partial write are considered droppable.
• Fix the broken influxd-ctl remove-data -force command.
• Fix security escalation bug in subscription management.
• Prevent certain user permissions from having a database-specific scope.
• Reduce the cost of the admin user check for clusters with large numbers of users.
• Fix hinted-handoff remote write batching.

v1.2.2 [2017-03-15]

This release builds off of the 1.2.1 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

Release Notes

Configuration Changes

The following configuration changes may need to changed before upgrading to 1.2.2 from prior versions.

shard-writer-timeout

We’ve removed the data node’s shard-writer-timeout configuration option from the [cluster] section. As of version 1.2.2, the system sets shard-writer-timeout internally. The configuration option can be removed from the data node configuration file.

retention-autocreate

In versions 1.2.0 and 1.2.1, the retention-autocreate setting appears in both the meta node and data node configuration files. To disable retention policy auto-creation, users on version 1.2.0 and 1.2.1 must set retention-autocreate to false in both the meta node and data node configuration files.

In version 1.2.2, we’ve removed the retention-autocreate setting from the data node configuration file. As of version 1.2.2, users may remove retention-autocreate from the data node configuration file. To disable retention policy auto-creation, set retention-autocreate to false in the meta node configuration file only.

This change only affects users who have disabled the retention-autocreate option and have installed version 1.2.0 or 1.2.1.

Bugfixes

Backup and Restore

• Prevent the shard not found error by making backups skip empty shards
• Prevent the shard not found error by making restore handle empty shards
• Ensure that restores from an incremental backup correctly handle file paths
• Allow incremental backups with restrictions (for example, they use the -db or rp flags) to be stores in the same directory
• Support restores on meta nodes that are not the raft leader

Hinted Handoff

• Fix issue where dropped writes were not recorded when the hinted handoff queue reached the maximum size
• Prevent the hinted handoff from becoming blocked if it encounters field type errors

Other

• Return partial results for the SHOW TAG VALUES query even if the cluster includes an unreachable data node
• Return partial results for the SHOW MEASUREMENTS query even if the cluster includes an unreachable data node
• Prevent a panic when the system files to process points
• Ensure that cluster hostnames can be case insensitive
• Update the retryCAS code to wait for a newer snapshot before retrying
• Serialize access to the meta client and meta store to prevent raft log buildup
• Remove sysvinit package dependency for RPM packages
• Make the default retention policy creation an atomic process instead of a two-step process
• Prevent influxd-ctl’s join argument from completing a join when the command also specifies the help flag (-h)
• Fix the influxd-ctl’s force removal of meta nodes
• Update the meta node and data node sample configuration files

v1.2.1 [2017-01-25]

Cluster-specific Bugfixes

• Fix panic: Slice bounds out of range  Fix how the system removes expired shards.
• Remove misplaced newlines from cluster logs

v1.2.0 [2017-01-24]

Release Notes

This release builds off of the 1.2.0 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

Upgrading

• The retention-autocreate configuration option has moved from the meta node configuration file to the data node configuration file. To disable the auto-creation of retention policies, set retention-autocreate to false in your data node configuration files.
• The previously deprecated influxd-ctl force-leave command has been removed. The replacement command to remove a meta node which is never coming back online is influxd-ctl remove-meta -force.

Cluster-specific Features

• Improve the meta store: any meta store changes are done via a compare and swap
• Add support for incremental backups
• Automatically remove any deleted shard groups from the data store
• Uncomment the section headers in the default configuration file
• Add InfluxQL support for subqueries

Cluster-specific Bugfixes

• Update dependencies with Godeps
• Fix a data race in meta client
• Ensure that the system removes the relevant user permissions and roles when a database is dropped
• Fix a couple typos in demo configuration file
• Make optional the version protobuf field for the meta store
• Remove the override of GOMAXPROCS
• Remove an unused configuration option (dir) from the backend
• Fix a panic around processing remote writes
• Return an error if a remote write has a field conflict
• Drop points in the hinted handoff that (1) have field conflict errors (2) have max-values-per-tag errors
• Remove the deprecated influxd-ctl force-leave command
• Fix issue where CQs would stop running if the first meta node in the cluster stops
• Fix logging in the meta httpd handler service
• Fix issue where subscriptions send duplicate data for Continuous Query results
• Fix the output for influxd-ctl show-shards
• Send the correct RPC response for ExecuteStatementRequestMessage

v1.1.5 [2017-04-28]

Bugfixes

• Prevent certain user permissions from having a database-specific scope.
• Fix security escalation bug in subscription management.

v1.1.3 [2017-02-27]

Release Notes

This release incorporates the changes in the 1.1.4 release of InfluxDB OSS. Please see the OSS changelog for more information about the OSS release.

Bugfixes

• Delay when a node listens for network connections until after all requisite services are running. This prevents queries to the cluster from failing unnecessarily.
• Allow users to set the GOMAXPROCS environment variable.

v1.1.2 [internal]

Release Notes

This release was an internal release only. It incorporates the changes in the 1.1.3 release of InfluxDB OSS. Please see the OSS changelog for more information about the OSS release.

v1.1.1 [2016-12-06]

Release Notes

This release builds off of the 1.1.1 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

This release is built with Go (golang) 1.7.4. It resolves a security vulnerability reported in Go (golang) version 1.7.3 which impacts all users currently running on the macOS platform, powered by the Darwin operating system.

Cluster-specific Bugfixes

• Fix hinted-handoff issue: Fix record size larger than max size  If a Hinted Handoff write appended a block that was larger than the maximum file size, the queue would get stuck because the maximum size was not updated. When reading the block back out during processing, the system would return an error because the block size was larger than the file size – which indicates a corrupted block.

v1.1.0 [2016-11-14]

Release Notes

This release builds off of the 1.1.0 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

Upgrading

• The 1.1.0 release of InfluxDB OSS has some important configuration changes that may affect existing clusters.
• The influxd-ctl join command has been renamed to influxd-ctl add-meta. If you have existing scripts that use influxd-ctl join, they will need to use influxd-ctl add-meta or be updated to use the new cluster setup command.

Cluster Setup

The influxd-ctl join command has been changed to simplify cluster setups. To join a node to a cluster, you can run influxd-ctl join <meta:8091>, and we will attempt to detect and add any meta or data node process running on the hosts automatically. The previous join command exists as add-meta now. If it’s the first node of a cluster, the meta address argument is optional.

Logging

Switches to journald logging for on systemd systems. Logs are no longer sent to /var/log/influxdb on systemd systems.

Cluster-specific Features

• Add a configuration option for setting gossiping frequency on data nodes
• Allow for detailed insight into the Hinted Handoff queue size by adding queueBytes to the hh_processor statistics
• Add authentication to the meta service API
• Update Go (golang) dependencies: Fix Go Vet and update circle Go Vet command
• Simplify the process for joining nodes to a cluster
• Include the node’s version number in the influxd-ctl show output
• Return and error if there are additional arguments after influxd-ctl show  Fixes any confusion between the correct command for showing detailed shard information (influxd-ctl show-shards) and the incorrect command (influxd-ctl show shards)

Cluster-specific Bugfixes

• Return an error if getting latest snapshot takes longer than 30 seconds
• Remove any expired shards from the /show-shards output
• Respect the pprof-enabled configuration setting and enable it by default on meta nodes
• Respect the pprof-enabled configuration setting on data nodes
• Use the data reference instead of Clone() during read-only operations for performance purposes
• Prevent the system from double-collecting cluster statistics
• Ensure that the meta API redirects to the cluster leader when it gets the ErrNotLeader error
• Don’t overwrite cluster users with existing InfluxDB OSS users when migrating an OSS instance into a cluster
• Fix a data race in the raft store
• Allow large segment files (> 10MB) in the Hinted Handoff
• Prevent copy-shard from retrying if the copy-shard command was killed
• Prevent a hanging influxd-ctl add-data command by making data nodes check for meta nodes before they join a cluster

v1.0.4 [2016-10-19]

Cluster-specific Bugfixes

• Respect the Hinted Handoff settings in the configuration file
• Fix expanding regular expressions when all shards do not exist on node that’s handling the request

v1.0.3 [2016-10-07]

Cluster-specific Bugfixes

• Fix a panic in the Hinted Handoff: lastModified

v1.0.2 [2016-10-06]

Release Notes

This release builds off of the 1.0.2 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

Cluster-specific Bugfixes

• Prevent double read-lock in the meta client
• Fix a panic around a corrupt block in Hinted Handoff
• Fix issue where systemctl enable would throw an error if the symlink already exists

v1.0.1 [2016-09-28]

Release Notes

This release builds off of the 1.0.1 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

Cluster-specific Bugfixes

• Balance shards correctly with a restore
• Fix a panic in the Hinted Handoff: runtime error: invalid memory address or nil pointer dereference
• Ensure meta node redirects to leader when removing data node
• Fix a panic in the Hinted Handoff: runtime error: makeslice: len out of range
• Update the data node configuration file so that only the minimum configuration options are uncommented

v1.0.0 [2016-09-07]

Release Notes

This release builds off of the 1.0.0 release of InfluxDB OSS. Please see the OSS release notes for more information about the OSS release.

Breaking Changes:

• The keywords IF, EXISTS, and NOT were removed for this release. This means you no longer need to specify IF NOT EXISTS for DROP DATABASE or IF EXISTS for CREATE DATABASE. Using these keywords will return a query error.
• max-series-per-database was added with a default of 1M but can be disabled by setting it to 0. Existing databases with series that exceed this limit will continue to load, but writes that would create new series will fail.

Hinted Handoff

A number of changes to hinted handoff are included in this release:

• Truncating only the corrupt block in a corrupted segment to minimize data loss
• Immediately queue writes in hinted handoff if there are still writes pending to prevent inconsistencies in shards
• Remove hinted handoff queues when data nodes are removed to eliminate manual cleanup tasks

Performance

• SHOW MEASUREMENTS and SHOW TAG VALUES have been optimized to work better for multiple nodes and shards
• DROP and DELETE statements run in parallel and more efficiently and should not leave the system in an inconsistent state

Security

The Cluster API used by influxd-ctl can not be protected with SSL certs.

Cluster Management

Data nodes that can no longer be restarted can now be forcefully removed from the cluster using influxd-ctl remove-data -force <addr>. This should only be run if a grace removal is not possible.

Backup and restore has been updated to fix issues and refine existing capabilities.

Cluster-specific Features

• Add the Users method to control client
• Add a -force option to the influxd-ctl remove-data command
• Disable the logging of stats service queries
• Optimize the SHOW MEASUREMENTS and SHOW TAG VALUES queries
• Update the Go (golang) package library dependencies
• Minimize the amount of data-loss in a corrupted Hinted Handoff file by truncating only the last corrupted segment instead of the entire file
• Log a write error when the Hinted Handoff queue is full for a node
• Remove Hinted Handoff queues on data nodes when the target data nodes are removed from the cluster
• Add unit testing around restore in the meta store
• Add full TLS support to the cluster API, including the use of self-signed certificates
• Improve backup/restore to allow for partial restores to a different cluster or to a database with a different database name
• Update the shard group creation logic to be balanced
• Keep raft log to a minimum to prevent replaying large raft logs on startup

Cluster-specific Bugfixes

• Remove bad connections from the meta executor connection pool
• Fix a panic in the meta store
• Fix a panic caused when a shard group is not found
• Fix a corrupted Hinted Handoff
• Ensure that any imported OSS admin users have all privileges in the cluster
• Ensure that max-select-series is respected
• Handle the peer already known error
• Fix Hinted handoff panic around segment size check
• Drop Hinted Handoff writes if they contain field type inconsistencies

Web Console

DEPRECATED: Enterprise Web Console

The Enterprise Web Console has officially been deprecated and will be eliminated entirely by the end of 2017. No additional features will be added and no additional bug fix releases are planned.

For browser-based access to InfluxDB Enterprise, Chronograf is now the recommended tool to use.