• Version 2.4.1 (2021-09-02)

Noteworthy Changes:

• Add a config flag to enable splitting queries across the rollup and
  raw data tables. Another config determines when to split the queries.
• Fix for CVE-2020-35476 that now validates and limits the inputs for
  Gnuplot query parameters to prevent remote code execution.
• Default log config will log CLI tools at INFO to standard out.
• New check_tsd_v2 script that evaluates individual metric groups given
  a filter.
• Collect stats from meta cache plugins.
• Add a python script to list and pretty-print queries running on a TSD.
• Add a single, standalone TSD systemd startup script and default to that
  instead of the multi-port TSD script.

Bug Fixes:

• Fix the "--sync" flag for FSCK to wait for repairs to execute against
  storage before proceeding.
• Fix expression queries to allow metric only and filterless queries.
• Fix an NPE in /api/query/last when appends are enabled.
• Fix races in the salt scanner and multigets on the storage maps.
• Fix rollup queries that need sum and count for downsampling instead of
  group bys.
• Fix fuzzy row filters for later versions of HBase by using filter pairs.
  And allow it to be combined with a regex filter.
• Fix stats from the individual salt scanners and overall query stats
  concurrency issues.
• Rename the stat "maxScannerUidtoStringTime" to
  "maxScannerUidToStringTime"
• Fix the min initial value for double values in the AggregationIterator
• Fix rollup queries producing unexpected results.
• Fix various UTs
• Support rollups in FSCK

• Version 2.4.0 (2018-12-16)

Noteworthy Changes:

• Set default data block encoding to DIFF in the create table script.
• Add callbacks to log errors in the FSCK tool when a call was made to
  fix something.
• Add a sum of squares aggregator "squareSum".
• Add the diff aggregator that computes the difference between the first
  and last values.
• Add a SystemD template to the RPM package.
• Allow tags to be added via HTTP header.
• Add example implementations for the Authorization and Authentication
  plugins.
• Change tsd.storage.use_otsdb_timestamp to default to false.
• Literal or filter now allows single character values.
• Rollup query code now only uses the downsampler value to pick an interval.
• Add jdk 8 in the debian script.
• Setup fill policies in the Nagios check

Bug Fixes:

• Fix rollup scanner filter for single aggregate queries.
• Fix FSCK HBase timestamps when deduping. Sometimes they were negative.
• Fix exception handling when writing data over HTTP with the sync flag enabled.
• Fix missing source files in the Makefile.
• Change UID cache to longs from ints and add hit and miss counters.
• Fix HighestCurrent returning the wrong results.
• Fix running query stats queryStart timestamp to millis.
• Fix TimeShift millisecond bug.
• Fix post remove step in the debian package.

• Version 2.3.2 (2018-12-16)

Noteworthy Changes:

• A new Python wrapper script to make FSCK repair runs easier.
• Track performance in the Nagios/Icinga script
• Add a Contributions file.
• Add a config, 'tsd.core.bulk.allow_out_of_order_timestamps' to allow out of
  order timestamps for bulk ingest.
• NOTE: This version also includes a JDK 8 compiled version of Jackson due to
  security patches. If you need to run with an older JDK please replace the
  Jackson JARs with older versions.

Bug Fixes:

• Unwrap NoSuchUniqueIds when writing data points to make it easier to understand
  exceptions.
• Fix an NPE in the PutDataPointRpc class if a data point in the list is null.
• Fix a Makefile error in the clean portion.
• Fix an NPOE in the UIDManager print result.
• Fix a bug in the UI where Y formats may contain a percent sign.
• Allow specifying the data block encoding and TTL in the HBase table creation
  script.
• Change the make and TSDB scripts to use relative paths.
• Fix parsing of use_meta from the URI for the search endpoint.
• Fix the clean cache script to be a bit more OS agnostic.

• Version 2.3.1 (2018-04-21)

Noteworthy Changes:

• When setting up aggregators, advance to the first data point equal to or greater
  than the query start timestamp. This helps with calendar downsampling intervals.
• Add support to the Nagios check script for downsampling fill policies.

Bug Fixes:

• Fix expression calculation by avoiding double execution and checking both
  output types for boolean values.
• Fixing missing tools scripts in builds.
• Default HBase 1.2.5 in the OSX install script
• Upgrade AsyncBigtable to 0.3.1
• Log query stats when a channel is closed unexpectedly.
• Add the Java 8 path in the debian init script and remove Java 6.
• Pass the column family name to the get requests in the compaction scheduler.
• Fix a comparison issue in the UI on group by tags.
• Filter annotation queries by the starting timestamp, excluding those in a row that
  began before the query start time.
• Tiny stap at purging backticks from Gnuplot scripts.
• Remove the final annotation from the meta classes so they can be extended.
• Fix the javacc maven plugin version.
• Fix the literal or filter to allow single character filters.
• Fix query start stats logging to use the ms instead of nano time.
• Move Jackson and Netty to newer versions for security reasons.
• Upgrade to AsyncHBase 1.8.2 for compatibility with HBase 1.3 and 2.0
• Fix the Highest Current calculation to handle empty time series.
• Change the cache hits counters to longs.

• Version 2.4.0 RC2 (2017-10-08)

Noteworthy Changes:

• Modify the RPC handler plugin system so that it parses only the first part of
  the URI instead of the entire path. Now plugins can implement sub-paths.
• Return the HTML 5 doctype for built-in UI pages
• Add an optional byte and/or data point limit to the amount of data fetched
  from storage. This allows admins to prevent OOMing TSDs due to massive queries.
• Allow a start time via config when enabling the date tiered compaction in HBase
• Provide the option of using an LRU for caching UIDs to avoid OOMing writers and
  readers with too many strings
• Optionally avoid writing to the forward or reverse UID maps when a specific TSD
  operational mode is enabled to avoid wasting memory on maps that will never be
  used.

Bug Fixes:

• Roll back UTF8 issue with UIDs in RC1 wherein the stored bytes weren't converting
  properly and vice-versa. We'll have to work on full UTF8 support in 3.x
• Fix a build issue for Javacc
• Add Kryo as a dependency to the fat jar
• Javadoc fixes
• Fix an issue with calendar aligned downsampling by seeking to the start time of
  the query when the zone-aligned timestamp may be earlier than the query start time
• Add the missing QueryLimitOverride to the makefile
• Fix compatibility with Bigtable for 2.4
• Enable standard read-only APIs when the TSD is in write only mode

• Version 2.3.0 (2016-12-31)

Noteworthy Changes:

• Release of 2.3.0.
• Add example classes for using the Java API.

Bug Fixes:

• Same fixes as in 2.2.2
• Fix a null UID check on decoding metric names from row keys.
• Fix unit tests for JDK 8 and later.

See also 2.3.0RC1 and 2.3.0RC2

• Version 2.2.2 (2016-12-29)

Bug Fixes:

• Fix an issue with writing metadata where using custom tags could cause the compare-
  and-set to fail due to variable ordering in Java's heap. Now tags are sorted so the
  custom tag ordering will be consistent.
• Fix millisecond queries that would miss data the top of the final hour if the end
  time was set to 1 second or less than the top of that final hour.
• Fix a concurrent modification issue where salt scanners were not synchronized on the
  annotation map and could cause spinning threads.

• Version 2.3.0 RC2 (2016-10-08)

Noteworthy Changes:

• Added a docker file and tool to build TSD docker containers (#871).
• Log X-Forwarded-For addresses when handling HTTP requests.
• Expand aggregator options in the Nagios check script.
• Allow enabling or disabling the HTTP API or UI.
• TSD will now exit when an unrecognized CLI param is passed.

Bug Fixes:

• Improved ALPN version detection when using Google Bigtable.
• Fix the DumpSeries class to support appended data point types.
• Fix queries where groupby is set to false on all filters.
• Fix a missing attribute in the Nagios check script (#728).
• Fix a major security bug where requesting a PNG with certain URI params could execute code
  on the host (#793, #781).
• Return a proper error code when dropping caches with the DELETE HTTP verb (#830).
• Fix backwards compatibility with HBase 0.94 when using explicit tags by removing the
  fuzzy filter (#837).
• Fix an RPM build issue when creating the GWT directory.

• Version 2.2.1 (2016-10-08)

Noteworthy Changes

• Generate an incrementing TSMeta request only if both enable_tsuid_incrementing and
  tsd.core.meta.enable_realtime_ts are enabled. Previously, increments would run
  regardless of whether or not the real time ts setting was enabled. If tsuid
  incrementing is disabled then a get and optional put is executed each time without
  modifying the meta counter field.
• Improve metadata storage performance by removing an extra getFromStorage() call.
• Add global Annotations to the gnuplot graphs (#773)
• Allow creation of a TSMeta object without a TSUID (#778)
• Move to AsyncHBase 1.7.2

Bug Fixes:

• Fix Python scripts to use the environment directory.
• Fix config name for "tsd.network.keep_alive" in included config files.
• Fix an issue with the filter metric and tag resolution chain during queries.
• Fix an issue with malformed, double dotted timestamps (#724).
• Fix an issue with tag filters where we need a copy before modifying the list.
• Fix comments in the config file around TCP no delay settings.
• Fix some query stats calculations around averaging and estimating the number
  of data points (#784).
• Clean out old .SWO files (#821)
• Fix a live-lock situation when performing regular expression or wildcard queries (#823).
• Change the static file path for the HTTP API to be relative (#857).
• Fix an issue where the GUI could flicker when two or more tag filters were set (#708).

• Version 2.3.0 RC1 (2016-05-02)

Noteworthy Changes:

• Introduced option --max-connection/tsd.core.connections.limit to set the maximum number
  of connection a TSD will accept (#638)
• 'tsdb import' can now read from stdin (#580)
• Added datapoints counter (#369)
• Improved metadata storage performance (#699)
• added checkbox for showing global annotations in UI (#736)
• Added startup plugins, can be used for Service Discovery or other integration (#719)
• Added MetaDataCache plugin api
• Added timeshift() function (#175)
• Now align downsampling to Gregorian Calendar (#548, #657)
• Added None aggregator to fetch raw data along with first and last aggregators to
  fetch only the first or last data points when downsampling.
• Added script to build OpenTSDB/HBase on OSX (#674)
• Add cross-series expressions with mathematical operators using Jexl
• Added query epxressions (alias(), scale(), absolute(), movingAverage(), highestCurrent(),
  highestMax(), timeShift(), divide(), sum(), difference(), multiply()) (#625)
• Add a Unique ID assignment filter API for enforcing UID assignment naming conventions.
• Add a whitelist regular expression based UID assignment filter
• Add a time series storage filter plugin API that allows processing time series data
  and determining if it should be stored or not.
• Allow using OpenTSDB with Google's Bigtable cloud platform or with Apache Cassandra

Bug Fixes:

• Some improperly formatted timestamps were allowed (#724)
• Removed stdout logging from packaged logback.xml files (#715)
• Restore the ability to create TSMeta objects via URI
• Restore raw data points (along with post-filtered data points) in query stats
• Built in UI will now properly display global annotations when the query string is passed