Lag Control v5

Commit scope kind: LAG CONTROL

Overview

Lag Control provides a mechanism where, if replication is running outside of limits set, a delay is injected into the origin node's client connections after processing transactions that make replicable updates. This delay is designed to slow the incoming transactions and bring replication back within the defined limits.

Background

The data throughput of database applications on a PGD origin node can exceed the rate at which committed data can replicate to downstream peer nodes.

If this imbalance persists, it can put satisfying organizational objectives such as RPO, RCO, and GEO at risk.

  • RPO (Recovery point objective) specifies the maximum-tolerated amount of data that can be lost due to unplanned events, usually expressed as an amount of time. In PGD, RPO determines the acceptable amount of committed data that hasn't been applied to one or more peer nodes.

  • RCO (Resource constraint objective) acknowledges that there is finite storage available. In PGD, the demands on these storage resources increase as lag increases.

  • GEO (Group elasticity objective) ensures that any node isn't originating new data at a rate that can't be saved to its peer nodes.

To allow organizations to achieve their objectives, PGD offers Lag Control. This feature provides a means to precisely regulate the potential imbalance without intruding on applications, by transparently introducing a delay to READ WRITE transactions that modify data. This delay, the PGD Commit Delay, starts at 0ms.

Using the LAG CONTROL commit scope kind, you can set a maximum time that commits commits can be delayed between nodes in a group, maximum lag time or maximum lag size (based on the size of the WAL).

If the nodes can process transactions within the specified maximums on enough nodes, the PGD commit delay will stay at 0ms or be reduced towards 0ms. If the maximums are exceeded on enough nodes though the PGD commit delay on the originating node is increased. It will continue increasing until the lag control constraints are met on enough nodes again.

The PGD commit delay happens after a transaction has completed and released all its locks and resources. This timing of the delay allows concurrent active transactions to carry on observing and modifying the delayed transactions values and acquiring its resources.

Strictly speaking, the PGD commit delay isn't a per-transaction delay. It's the mean value of commit delays over a stream of transactions for a particular client connection. This technique allows the commit delay and fine-grained adjustments of the value to escape the coarse granularity of OS schedulers, clock interrupts, and variation due to system load. It also allows the PGD runtime commit delay to settle within microseconds of the lowest duration possible to maintain a lag measure threshold.

PGD commit delay != Postgres commit delay

Don't conflate the PGD commit delay with the Postgres commit delay. They are unrelated and perform different functions. Don't substitute one for the other.

Requirements

To get started using Lag Control:

  • Determine the maximum acceptable commit delay time max_commit_delay that all database applications can tolerate.

  • Decide on the lag measure to use. Choose either lag size max_lag_size or lag time max_lag_time.

  • Decide on the groups or subgroups involved and the minimum number of nodes in each collection required to satisfy confirmation. This information forms the basis for the definition of a commit scope rule.

Configuration

You specify lag control in a commit scope, which allows consistent and coordinated parameter settings across the nodes spanned by the commit scope rule. You can include a lag control specification in the default commit scope of a top group or as part of an origin group commit scope.

As in example, take a configuration with two datacenters, left_dc and right_dc, represented as sub-groups:

SELECT bdr.create_node_group(
    node_group_name := 'left_dc',
    parent_group_name := 'top_group',
    join_node_group := false
);
SELECT bdr.create_node_group(
    node_group_name := 'right_dc',
    parent_group_name := 'top_group',
    join_node_group := false
);

The code below adds Lag Control rules for those two data centers, using individual rules for each subgroup:

SELECT bdr.add_commit_scope(
    commit_scope_name := 'example_scope',
    origin_node_group := 'left_dc',
    rule := 'ALL (left_dc) LAG CONTROL (max_commit_delay=500ms, max_lag_time=30s) AND ANY 1 (right_dc) LAG CONTROL (max_commit_delay=500ms, max_lag_time=30s)',
    wait_for_ready := true
);
SELECT bdr.add_commit_scope(
    commit_scope_name := 'example_scope',
    origin_node_group := 'right_dc',
    rule := 'ANY 1 (left_dc) LAG CONTROL (max_commit_delay=0.250ms, max_lag_size=100MB) AND ALL (right_dc) LAG CONTROL (max_commit_delay=0.250ms, max_lag_size=100MB)',
    wait_for_ready := true
);

You can add a lag control commit scope rule to existing commit scope rules that also include group commit and CAMO rule specifications.

The max_commit_delay is interval, typically specified in milliseconds (1ms). Using fractional values for sub-millisecond precision is supported.

The max_lag_size is an integer which specifies the maximum allowed lag in terms of WAL bytes.

The max_lag_time is an interval, typically specified in seconds, that specifies the maximum allowed lag in terms of time.

The maximum commit delay (max_commit_delay) is a ceiling value representing a hard limit, which means that a commit delay never exceeds the configured value.

The maximum lag size and time (max_lag_size and max_lag_time) are soft limits that can be exceeded. When the maximum commit delay is reached, there's no additional back pressure on the lag measures to prevent their continued increase.

Confirmation

Confirmation LevelLag Control Handling
receivedNot applicable, only uses the default VISIBLE.
replicatedNot applicable, only uses the default VISIBLE.
durableNot applicable, only uses the default VISIBLE.
visible (default)Not applicable, only uses the default VISIBLE.

Transaction application

The PGD commit delay is applied to all READ WRITE transactions that modify data for user applications. This behavior implies that any transaction that doesn't modify data, including declared READ WRITE transactions, is exempt from the commit delay.

Asynchronous transaction commit also executes a PGD commit delay. This might appear counterintuitive, but asynchronous commit, by virtue of its performance, can be one of the greatest sources of replication lag.

Postgres and PGD auxillary processes don't delay at transaction commit. Most notably, PGD writers don't execute a commit delay when applying remote transactions on the local node. This is by design, as PGD writers contribute nothing to outgoing replication lag and can reduce incoming replication lag the most by not having their transaction commits throttled by a delay.

Limitations

The maximum commit delay is a ceiling value representing a hard limit, which means that a commit delay never exceeds the configured value. Conversely, the maximum lag measures both by size and time and are soft limits that can be exceeded. When the maximum commit delay is reached, there's no additional back pressure on the lag measures to prevent their continued increase.

There's no way to exempt origin transactions that don't modify PGD replication sets from the commit delay. For these transactions, it can be useful to SET LOCAL the maximum transaction delay to 0.

Caveats

Application TPS is one of many factors that can affect replication lag. Other factors include the average size of transactions for which PGD commit delay can be less effective. In particular, bulk load operations can cause replication lag to rise, which can trigger a concomitant rise in the PGD runtime commit delay beyond the level reasonably expected by normal applications, although still under the maximum allowed delay.

Similarly, an application with a very high OLTP requirement and modest data changes can be unduly restrained by the acceptable PGD commit delay setting.

In these cases, it can be useful to use the SET [SESSION|LOCAL] command to custom configure Lag Control settings for those applications or modify those applications. For example, bulk load operations are sometimes split into multiple, smaller transactions to limit transaction snapshot duration and WAL retention size or establish a restart point if the bulk load fails. In deference to Lag Control, those transaction commits can also schedule very long PGD commit delays to allow digestion of the lag contributed by the prior partial bulk load.

Meeting organizational objectives

In the example objectives list earlier:

  • RPO can be met by setting an appropriate maximum lag time.
  • RCO can be met by setting an appropriate maximum lag size.
  • GEO can be met by monitoring the PGD runtime commit delay and the PGD runtime lag measures,

As mentioned, when the maximum PGD runtime commit delay is pegged at the PGD configured commit-delay limit, and the lag measures consistently exceed their PGD-configured maximum levels, this scenario can be a marker for PGD group expansion.

Lag Control and Extensions

The PGD commit delay is a post-commit delay. It occurs after the transaction has committed and after all Postgres resources locked or acquired by the transaction are released. Therefore, the delay doesn't prevent concurrent active transactions from observing or modifying its values or acquiring its resources. The same guarantee can't be made for external resources managed by Postgres extensions. Regardless of extension dependencies, the same guarantee can be made if the PGD extension is listed before extension-based resource managers in postgresql.conf.