v15.0.0
版本发布时间: 2022-10-25 23:35:42
vitessio/vitess最新发布版本:v21.0.0-rc2(2024-10-23 13:13:47)
Release of Vitess v15.0.0
Summary
- Known Issues
- Breaking Changes
- Deprecations
- Command-Line Syntax Deprecations
- VTGate
- VDiff2
- New command line flags and behavior
- Online DDL Changes
- Tablet Throttler
- Mysql Compatibility
- Durability Policy
- New EXPLAIN Format
- VTOrc
- Flags Restructure
Known Issues
- Corrupted results for non-full-group-by queries with JOINs. This can be resolved by using full-group-by queries.
Major Changes
Breaking Changes
Flags
- The deprecated
--cpu_profile
flag has been removed. Please use the--pprof
flag instead. - The deprecated
--mem-profile-rate
flag has been removed. Please use--pprof=mem
instead. - The deprecated
--mutex-profile-fraction
flag has been removed. Please use--pprof=mutex
instead. - The deprecated vtgate/vtexplain/vtcombo flag
--planner_version
has been removed. Please use--planner-version
instead. - The deprecated flag
--master_connect_retry
has been removed. Please use--replication_connect_retry
instead. -
vtctl
commands that take shard names and ranges as positional arguments (e.g.vtctl Reshard ks.workflow -80 -40,40-80
) need to have their positional arguments separated from their flag arguments by a double-dash separator to avoid the new parsing library from mistaking them as flags (e.g.vtctl Reshard ks.workflow -- -80 -40,40-80
). - The
--cell
flag in thevtgate
binary no longer has a default value. It is a required argument that has to be specified for the binary to run. Please explicitly specify the flag, if dependent on the flag's default value. - The
--db-config-*-*
VTTablet flags were deprecated inv3.0.0
. They have now been deleted as part of this release. You must use--db_dba_*
now.
vttablet Flag Deletions
The following VTTablet flags were deprecated in 7.0. They have now been deleted
-
--queryserver-config-message-conn-pool-size
-
--queryserver-config-message-conn-pool-prefill-parallelism
-
--client-found-rows-pool-size
A different existing flag--queryserver-config-transaction-cap
will be used instead -
--transaction_shutdown_grace_period
Use--shutdown_grace_period
instead -
--queryserver-config-max-dml-rows
-
--queryserver-config-allowunsafe-dmls
-
--pool-name-prefix
-
--enable-autocommit
Autocommit is always allowed
Vindex Interface
All the vindex interface methods are changed by adding context.Context
as an input parameter.
E.g:
Map(vcursor VCursor, .... ) ....
To
Map(ctx context.Context, vcursor VCursor, .... ) ....
This only affects users who have added their own custom vindex implementation. They are required to change their implementation with these new interface method expectations.
Deprecations
LogStats Table and Keyspace deprecated
Information about which tables are used was being reported through the Keyspace
and Table
fields on LogStats.
For multi-table queries, this output can be confusing, so we have added TablesUsed
, that is a string array, listing all tables and which keyspace they are on.
Keyspace
and Table
fields are deprecated and will be removed in the v16 release of Vitess.
Orchestrator Integration Deprecation
Orchestrator integration in vttablet
has been deprecated. It will continue to work in this release but is liable to be removed in future releases.
Consider using VTOrc instead of Orchestrator as VTOrc goes GA in this release.
Connection Pool Prefill
The connection pool with prefilled connections have been removed. The pool now does lazy connection creation.
InitShardPrimary Deprecation
The vtcltd command InitShardPrimary has been deprecated. Please use PlannedReparentShard instead.
Command-line syntax deprecations
vttablet startup flag deletions
The following VTTablet flags were deprecated in 7.0. They have now been deleted
- --queryserver-config-message-conn-pool-size
- --queryserver-config-message-conn-pool-prefill-parallelism
- --client-found-rows-pool-size --queryserver-config-transaction-cap will be used instead
- --transaction_shutdown_grace_period Use --shutdown_grace_period instead
- --queryserver-config-max-dml-rows
- --queryserver-config-allowunsafe-dmls
- --pool-name-prefix
- --enable-autocommit Autocommit is always allowed
vttablet startup flag deprecations
-
--enable-query-plan-field-caching
has been deprecated. It will be removed in v16. -
--enable_semi_sync
has been deprecated. It will be removed in v16. Instead, set the correct durability policy usingSetKeyspaceDurabilityPolicy
-
--queryserver-config-pool-prefill-parallelism
,--queryserver-config-stream-pool-prefill-parallelism
and--queryserver-config-transaction-prefill-parallelism
have all been deprecated. They will be removed in v16. -
--backup_storage_hook
has been deprecated, consider using one of the builtin compression algorithms or--external-compressor
and--external-decompressor
instead.
vtbackup flag deprecations
-
--backup_storage_hook
has been deprecated, consider using one of the builtin compression algorithms or--external-compressor
and--external-decompressor
instead.
VTGate
vtgate --mysql-server-pool-conn-read-buffers
--mysql-server-pool-conn-read-buffers
enables pooling of buffers used to read from incoming
connections, similar to the way pooling happens for write buffers. Defaults to off.
VDiff v2
Resume Workflow
We introduced the ability to resume a VDiff2 workflow:
$ vtctlclient --server=localhost:15999 VDiff --v2 customer.commerce2customer resume 4c664dc2-eba9-11ec-9ef7-920702940ee0
VDiff 4c664dc2-eba9-11ec-9ef7-920702940ee0 resumed on target shards, use show to view progress
$ vtctlclient --server=localhost:15999 VDiff --v2 customer.commerce2customer show last
VDiff Summary for customer.commerce2customer (4c664dc2-eba9-11ec-9ef7-920702940ee0)
State: completed
RowsCompared: 196
HasMismatch: false
StartedAt: 2022-06-26 22:44:29
CompletedAt: 2022-06-26 22:44:31
Use "--format=json" for more detailed output.
$ vtctlclient --server=localhost:15999 VDiff --v2 --format=json customer.commerce2customer show last
{
"Workflow": "commerce2customer",
"Keyspace": "customer",
"State": "completed",
"UUID": "4c664dc2-eba9-11ec-9ef7-920702940ee0",
"RowsCompared": 196,
"HasMismatch": false,
"Shards": "0",
"StartedAt": "2022-06-26 22:44:29",
"CompletedAt": "2022-06-26 22:44:31"
}
We leverage this resume capability to automatically restart a VDiff2 workflow that encountered a retryable error.
We also made a number of other enhancements like progress reporting and features needed to make it a full replacement for VDiff v1. You can see more details in the tracking ticket for the VDiff2 feature complete target: https://github.com/vitessio/vitess/issues/10494
Now that VDiff v2 is feature complete in 15.0, we hope to make it GA in 16.0.
Please see the VDiff2 documentation for additional information.
New command line flags and behavior
vtctl GetSchema --table-schema-only
The new flag --table-schema-only
skips column introspection. GetSchema
only returns general schema analysis, and specifically it includes the CREATE TABLE|VIEW
statement in the schema
field.
Support for additional compressors and decompressors during backup & restore
Backup/Restore now allow you many more options for compression and decompression instead of relying on the default compressor(pargzip
).
There are some built-in compressors which you can use out-of-the-box. Users will need to evaluate which option works best for their
use-case. Here are the flags that control this feature
-
--compression-engine-name
-
--external-compressor
-
--external-decompressor
-
--external-compressor-extension
-
--compression-level
--compression-engine-name
specifies the engine used for compression. It can have one of the following values
- pargzip (Default)
- pgzip
- lz4
- zstd
- external
If you want to use any of the built-in compressors, simply set one of the above values other than external
for --compression-engine-name
. The value
specified in --compression-engine-name
is saved in the backup MANIFEST, which is later read by the restore process to decide which
engine to use for decompression. Default value for engine is 'pargzip'.
If you would like to use a custom command or external tool for compression/decompression then you need to provide the full command with
arguments to the --external-compressor
and --external-decompressor
flags. --external-compressor-extension
flag also needs to be provided
so that compressed files are created with the correct extension. If the external command is not using any of the built-in compression engines
(i.e. pgzip, pargzip, lz4 or zstd) then you need to set --compression-engine-name
to value 'external'.
Please note that if you want to keep the current behavior then you don't need to provide any of these flags. You can read more about backup & restore [here] (https://vitess.io/docs/15.0/user-guides/operating-vitess/backup-and-restore/).
If you decided to switch from an external compressor to one of the built-in supported compressors (i.e. pgzip, pargzip, lz4 or zstd) at any point in the future, you will need to do it in two steps.
- step #1, set
--external-compressor
and--external-compressor-extension
flag values to empty and change--compression-engine-name
to desired value. - Step #2, after at least one cycle of backup with new configuration, you can set
--external-decompressor
flag value to empty.
The reason you cannot change all the values together is because the restore process will then have no way to find out which external decompressor should be used to process the previous backup. Please make sure you have thought out all possible scenarios for restore before transitioning from one compression engine to another.
Independent OLAP and OLTP transactional timeouts
--queryserver-config-olap-transaction-timeout
specifies the timeout applied
to a transaction created within an OLAP workload. The default value is 30
seconds, but this can be raised, lowered, or set to zero to disable the timeout
altogether.
Until now, while OLAP queries would bypass the query timeout, transactions
created within an OLAP session would be rolled back
--queryserver-config-transaction-timeout
seconds after the transaction was
started.
As of now, OLTP and OLAP transaction timeouts can be configured independently of each other.
The main use case is to run queries spanning a long period of time which require transactional guarantees such as consistency or atomicity.
Support for specifying group information in calls to VTGate
--grpc-use-effective-groups
allows non-SSL callers to specify groups information for a caller.
Until now, you could only specify the caller-id for the security context used to authorize queries.
As of now, you can specify the principal of the caller, and any groups they belong to.
Online DDL changes
Concurrent vitess migrations
All Online DDL migrations using the vitess
strategy are now eligible to run concurrently, given --allow-concurrent
DDL strategy flag. Until now, only CREATE
, DROP
and REVERT
migrations were eligible, and now ALTER
migrations are supported, as well. The terms for ALTER
migrations concurrency:
- DDL strategy must be
vitess --allow-concurent ...
- No two migrations can run concurrently on the same table
- No two
ALTER
s will copy table data concurrently - A concurrent
ALTER
migration will not start if anotherALTER
is running and is notready_to_complete
The main use case is to run multiple concurrent migrations, all with --postpone-completion
. All table-copy operations will run sequentially, but no migration will actually cut-over, and eventually all migrations will be ready_to_complete
, continuously tailing the binary logs and keeping up-to-date. A quick and iterative ALTER VITESS_MIGRATION '...' COMPLETE
sequence of commands will cut-over all migrations closely together (though not atomically together).
vtctl command changes.
All online DDL show
commands can now be run with a few additional parameters
-
--order
, order migrations in the output by either ascending or descending order of theirid
fields. -
--skip
, skip specified number of migrations in the output. -
--limit
, limit results to a specified number of migrations in the output.
New syntax
The following is now supported:
ALTER VITESS_MIGRATION COMPLETE ALL
This works on all pending migrations (queued
, ready
, running
) and internally issues a ALTER VITESS_MIGRATION '<uuid>' COMPLETE
for each one. The command is useful for completing multiple concurrent migrations (see above) that are open-ended (--postpone-completion
).
Tablet Throttler
API changes
API endpoint /debug/vars
now exposes throttler metrics, such as number of hits and errors per app per check type. Example:
$ curl -s http://127.0.0.1:15100/debug/vars | jq . | grep Throttler
"ThrottlerAggregatedMysqlSelf": 0.191718,
"ThrottlerAggregatedMysqlShard": 0.960054,
"ThrottlerCheckAnyError": 27,
"ThrottlerCheckAnyMysqlSelfError": 13,
"ThrottlerCheckAnyMysqlSelfTotal": 38,
"ThrottlerCheckAnyMysqlShardError": 14,
"ThrottlerCheckAnyMysqlShardTotal": 42,
"ThrottlerCheckAnyTotal": 80,
"ThrottlerCheckMysqlSelfSecondsSinceHealthy": 0,
"ThrottlerCheckMysqlShardSecondsSinceHealthy": 0,
"ThrottlerProbesLatency": 355523,
"ThrottlerProbesTotal": 74,
Mysql Compatibility
System Settings
Vitess has had support for system settings from release 7.0 onwards, but this support came with some caveats. As soon as a client session changes a default system setting, a mysql connection gets reserved for it. This can sometimes lead to clients running out of mysql connections. Users were instructed to minimize the use of this feature and to try to set the desired system settings as defaults in the mysql config.
With this release, Vitess can handle system settings changes in a much better way and clients can use them more freely. Vitess now has the ability to pool changed settings without reserving connections for any particular session.
This feature can be enabled by setting queryserver-enable-settings-pool
flag on the vttablet. It is disabled by default.
In future releases, we will make this flag enabled by default.
Lookup Vindexes
Lookup vindexes now support a new parameter multi_shard_autocommit
. If this is set to true
, lookup vindex dml queries will be sent as autocommit to all shards instead of being wrapped in a transaction.
This is different from the existing autocommit
parameter where the query is sent in its own transaction separate from the ongoing transaction if any i.e. begin -> lookup query execs -> commit/rollback
Durability Policy
Cross Cell
A new durability policy cross_cell
is now supported. cross_cell
durability policy only allows replica tablets from a different cell than the current primary to
send semi-sync ACKs. This ensures that any committed write exists in at least 2 tablets belonging to different cells.
New EXPLAIN format
FORMAT=vtexplain
With this new explain
format, you can get an output that is very similar to the command line vtexplain
app, but from a running vtgate
, through a MySQL query.
VTOrc
Old UI Removal and Replacement
The old UI that VTOrc inherited from Orchestrator
has been removed. A replacement UI, more consistent with the other Vitess binaries has been created.
In order to use the new UI, --port
flag has to be provided.
Along with the UI, the old APIs have also been deprecated. However, some of them have been ported over to the new UI -
Old API | New API | Additional notes |
---|---|---|
/api/problems |
/api/problems |
The new API also supports filtering using the keyspace and shard name |
/api/disable-global-recoveries |
/api/disable-global-recoveries |
Functionally remains the same |
/api/enable-global-recoveries |
/api/enable-global-recoveries |
Functionally remains the same |
/api/health |
/debug/health |
Functionally remains the same |
/api/replication-analysis |
/api/replication-analysis |
Functionally remains the same. Output is now JSON format. |
Apart from these APIs, we also now have /debug/status
, /debug/vars
and /debug/liveness
available in the new UI.
Configuration Refactor and New Flags
Since VTOrc was forked from Orchestrator
, it inherited a lot of configurations that don't make sense for the Vitess use-case.
All of such configurations have been removed.
VTOrc ignores the configurations that it doesn't understand. So old configurations can be kept around on upgrading and won't cause any issues. They will just be ignored.
For all the configurations that are kept, flags have been added for them and the flags are the desired way to pass these configurations going forward. The config file will be deprecated and removed in upcoming releases. The following is a list of all the configurations that are kept and the associated flags added.
Configurations Kept | Flags Introduced |
---|---|
SQLite3DataFile | --sqlite-data-file |
InstancePollSeconds | --instance-poll-time |
SnapshotTopologiesIntervalHours | --snapshot-topology-interval |
ReasonableReplicationLagSeconds | --reasonable-replication-lag |
AuditLogFile | --audit-file-location |
AuditToSyslog | --audit-to-backend |
AuditToBackendDB | --audit-to-syslog |
AuditPurgeDays | --audit-purge-duration |
RecoveryPeriodBlockSeconds | --recovery-period-block-duration |
PreventCrossDataCenterPrimaryFailover | --prevent-cross-cell-failover |
LockShardTimeoutSeconds | --lock-shard-timeout |
WaitReplicasTimeoutSeconds | --wait-replicas-timeout |
TopoInformationRefreshSeconds | --topo-information-refresh-duration |
RecoveryPollSeconds | --recovery-poll-duration |
Apart from configurations, some flags from VTOrc have also been removed -
-
sibling
-
destination
-
discovery
-
skip-unresolve
-
skip-unresolve-check
-
noop
-
binlog
-
statement
-
grab-election
-
promotion-rule
-
skip-continuous-registration
-
enable-database-update
-
ignore-raft-setup
-
tag
The ideal way to ensure backward compatibility is to remove the flags listed above while on the previous release. Then upgrade VTOrc. After upgrading, remove the config file and instead pass the flags that are introduced.
Example Upgrade
If you are running VTOrc with the flags --ignore-raft-setup --clusters_to_watch="ks/0" --config="path/to/config"
and the following configuration
{
"Debug": true,
"ListenAddress": ":6922",
"MySQLTopologyUser": "orc_client_user",
"MySQLTopologyPassword": "orc_client_user_password",
"MySQLReplicaUser": "vt_repl",
"MySQLReplicaPassword": "",
"RecoveryPeriodBlockSeconds": 1,
"InstancePollSeconds": 1,
"PreventCrossDataCenterPrimaryFailover": true
}
First drop the flag --ignore-raft-setup
while on the previous release. So, you'll be running VTOrc with --clusters_to_watch="ks/0" --config="path/to/config"
and the same configuration listed above.
Now you can upgrade your VTOrc version continuing to use the same flags and configurations, and it will continue to work just the same. If you wish to use the new UI, then you can add the --port
flag as well.
After upgrading, you can drop the configuration entirely and use the new flags like --clusters_to_watch="ks/0" --recovery-period-block-duration=1s --instance-poll-time=1s --prevent-cross-cell-failover
Default Configuration Files
The default files that VTOrc searches for configurations in have also changed from "/etc/orchestrator.conf.json", "conf/orchestrator.conf.json", "orchestrator.conf.json"
to
"/etc/vtorc.conf.json", "conf/vtorc.conf.json", "vtorc.conf.json"
.
Flags Restructure
Flags Diff
In addition to these major streams of work in release-15.0, we have made tremendous progress on VEP-4, aka The Flag Situation, reorganizing our code so that Vitess binaries and their flags are clearly aligned in help text. An immediate win for usability, this positions us well to move on to a viper implementation which will facilitate additional improvements including standardization of flag syntax and runtime configuration reloads. We are also aligning with industry standards regarding the use of flags, ensuring a seamless experience for users migrating from or integrating with other platforms. Below are the changes for each binary.
- mysqlctl
- mysqlctld
- vtaclcheck
- vtadmin
- vtctlclient
- vtctld
- vtctldclient
- vtexplain
- vtgate
- vtgtr
- vtorc
- vttablet
- vttestserver
- vttlstest
- zk
- zkctl
- zkctld
The entire changelog for this release can be found here.
The release includes 595 commits (excluding merges)
Thanks to all our contributors: @Abirdcfly, @DeathBorn, @GuptaManan100, @K-Kumar-01, @L3o-pold, @Phanatic, @Weijun-H, @ajm188, @arthurschreiber, @arvind-murty, @brirams, @dbussink, @deepthi, @dependabot[bot], @doeg, @frouioui, @harshit-gangal, @mattlord, @maxenglander, @mgale, @notfelineit, @ofiriluz, @olyazavr, @quinox, @rafer, @renatolabs, @rohit-nayak-ps, @rsajwani, @rvrangel, @saunderst, @shlomi-noach, @systay, @vitess-bot[bot], @vmg, @yoheimuta
1、 vitess-15.0.0-e8c7e27.tar.gz 453.54MB
2、 vitess-15.0.0-e8c7e27.x86_64.rpm 451.92MB
3、 vitess_15.0.0-e8c7e27_amd64.deb 453.55MB