Re: Incremental View Maintenance, take 2

On Thu, 1 Jun 2023 23:59:09 +0900
Yugo NAGATA <nagata(at)sraoss(dot)co(dot)jp> wrote:

> Hello hackers,
>
> Here's a rebased version of the patch-set adding Incremental View
> Maintenance support for PostgreSQL. That was discussed in [1].

> [1] https://www.postgresql.org/message-id/flat/20181227215726.4d166b4874f8983a641123f5%40sraoss.co.jp

---------------------------------------------------------------------------------------
* Overview

Incremental View Maintenance (IVM) is a way to make materialized views
up-to-date by computing only incremental changes and applying them on
views. IVM is more efficient than REFRESH MATERIALIZED VIEW when
only small parts of the view are changed.

** Feature

The attached patchset provides a feature that allows materialized views
to be updated automatically and incrementally just after a underlying
table is modified.

You can create an incementally maintainable materialized view (IMMV)
by using CREATE INCREMENTAL MATERIALIZED VIEW command.

The followings are supported in view definition queries:
- SELECT ... FROM ... WHERE ..., joins (inner joins, self-joins)
- some built-in aggregate functions (count, sum, avg, min, max)
- GROUP BY clause
- DISTINCT clause

Views can contain multiple tuples with the same content (duplicate tuples).

** Restriction

The following are not supported in a view definition:
- Outer joins
- Aggregates otehr than above, window functions, HAVING
- Sub-queries, CTEs
- Set operations (UNION, INTERSECT, EXCEPT)
- DISTINCT ON, ORDER BY, LIMIT, OFFSET

Also, a view definition query cannot contain other views, materialized views,
foreign tables, partitioned tables, partitions, VALUES, non-immutable functions,
system columns, or expressions that contains aggregates.

---------------------------------------------------------------------------------------
* Design

An IMMV is maintained using statement-level AFTER triggers.
When an IMMV is created, triggers are automatically created on all base
tables contained in the view definition query.

When a table is modified, changes that occurred in the table are extracted
as transition tables in the AFTER triggers. Then, changes that will occur in
the view are calculated by a rewritten view dequery in which the modified table
is replaced with the transition table.

For example, if the view is defined as "SELECT * FROM R, S", and tuples inserted
into R are stored in a transiton table dR, the tuples that will be inserted into
the view are calculated as the result of "SELECT * FROM dR, S".

** Multiple Tables Modification

Multiple tables can be modified in a statement when using triggers, foreign key
constraint, or modifying CTEs. When multiple tables are modified, we need
the state of tables before the modification.

For example, when some tuples, dR and dS, are inserted into R and S respectively,
the tuples that will be inserted into the view are calculated by the following
two queries:

"SELECT * FROM dR, S_pre"
"SELECT * FROM R, dS"

where S_pre is the table before the modification, R is the current state of
table, that is, after the modification. This pre-update states of table
is calculated by filtering inserted tuples and appending deleted tuples.
The subquery that represents pre-update state is generated in get_prestate_rte().
Specifically, the insterted tuples are filtered by calling IVM_visible_in_prestate()
in WHERE clause. This function checks the visibility of tuples by using
the snapshot taken before table modification. The deleted tuples are contained
in the old transition table, and this table is appended using UNION ALL.

Transition tables for each modification are collected in each AFTER trigger
function call. Then, the view maintenance is performed in the last call of
the trigger.

In the original PostgreSQL, tuplestores of transition tables are freed at the
end of each nested query. However, their lifespan needs to be prolonged to
the end of the out-most query in order to maintain the view in the last AFTER
trigger. For this purpose, SetTransitionTablePreserved is added in trigger.c.

** Duplicate Tulpes

When calculating changes that will occur in the view (= delta tables),
multiplicity of tuples are calculated by using count(*).

When deleting tuples from the view, tuples to be deleted are identified by
joining the delta table with the view, and tuples are deleted as many as
specified multiplicity by numbered using row_number() function.
This is implemented in apply_old_delta().

When inserting tuples into the view, each tuple is duplicated to the
specified multiplicity using generate_series() function. This is implemented
in apply_new_delta().

** DISTINCT clause

When DISTINCT is used, the view has a hidden column __ivm_count__ that
stores multiplicity for tuples. When tuples are deleted from or inserted into
the view, the values of __ivm_count__ column is decreased or increased as many
as specified multiplicity. Eventually, when the values becomes zero, the
corresponding tuple is deleted from the view. This is implemented in
apply_old_delta_with_count() and apply_new_delta_with_count().

** Aggregates

Built-in count sum, avg, min, and max are supported. Whether a given
aggregate function can be used or not is checked by using its OID in
check_aggregate_supports_ivm().

When creating a materialized view containing aggregates, in addition
to __ivm_count__, more than one hidden columns for each aggregate are
added to the target list. For example, columns for storing sum(x),
count(x) are added if we have avg(x). When the view is maintained,
aggregated values are updated using these hidden columns, also hidden
columns are updated at the same time.

The maintenance of aggregated view is performed in
apply_old_delta_with_count() and apply_new_delta_with_count(). The SET
clauses for updating columns are generated by append_set_clause_*().

If the view has min(x) or max(x) and the minimum or maximal value is
deleted from a table, we need to update the value to the new min/max
recalculated from the tables rather than incremental computation. This
is performed in recalc_and_set_values().

---------------------------------------------------------------------------------------
* Details of the patch-set (v28)

> The patch-set consists of the following eleven patches.

In the previous version, the number of patches were nine.
In the latest patch-set, the patches are divided more finely
aiming to make the review easier.

> - 0001: Add a syntax to create Incrementally Maintainable Materialized Views

The prposed syntax to create an incrementally maintainable materialized
view (IMMV) is;

CREATE INCREMENTAL MATERIALIZED VIEW AS SELECT .....;

However, this syntax is tentative, so any suggestions are welcomed.

> - 0002: Add relisivm column to pg_class system catalog

We add a new field in pg_class to indicate a relation is IMMV.
Another alternative is to add a new catalog for managing materialized
views including IMMV, but I am not sure if we want this.

> - 0003: Allow to prolong life span of transition tables until transaction end

This patch fixes the trigger system to allow to prolong lifespan of
tuple stores for transition tables until the transaction end. We need
this because multiple transition tables have to be preserved until the
end of the out-most query when multiple tables are modified by nested
triggers. (as explained above in Design - Multiple Tables Modification)

If we don't want to change the trigger system in such way, the alternative
is to copy the contents of transition tables to other tuplestores, although
it needs more time and memory.

> - 0004: Add Incremental View Maintenance support to pg_dump

This patch enables pg_dump to output IMMV using the new syntax.

> - 0005: Add Incremental View Maintenance support to psql

This patch implements tab-completion for the new syntax and adds
information of IMMV to \d meta-command results.

> - 0006: Add Incremental View Maintenance support

This patch implements the basic IVM feature.
DISTINCT and aggregate are not supported here.

When an IMMV is created, the view query is checked, and if any
non-supported feature is used, it raises an error. If it is ok,
triggers are created on base tables and an unique index is
created on the view if possible.

In BEFORE trigger, an entry is created for each IMMV and the number
of trigger firing is counted. Also, the snapshot just before the
table modification is stored.

In AFTER triggers, each transition tables are preserved. The number
of trigger firing is counted also here, and when the firing number of
BEFORE and AFTER trigger reach the same, it is deemed the final AFTER
trigger call.

In the final AFTER trigger, the IMMV is maintained. Rewritten view
query is executed to generate delta tables, and deltas are applied
to the view. If multiple tables are modified simultaneously, this
process is iterated for each modified table. Tables before processed
are represented in "pre-update-state", processed tables are
"post-update-state" in the rewritten query.

> - 0007: Add DISTINCT support for IVM

This patch adds DISTINCT clause support.

When an IMMV including DISTINCT is created, a hidden column
"__ivm_count__" is added to the target list. This column has the
number of duplicity of the same tuples. The duplicity is calculated
by adding "count(*)" and GROUP BY to the view query.

When an IMMV is maintained, the duplicity in __ivm_count__ is updated,
and a tuples whose duplicity becomes zero can be deleted from the view.
This logic is implemented by SQL in apply_old_delta_with_count and
apply_new_delta_with_count.

Columns starting with "__ivm_" are deemed hidden columns that doesn't
appear when a view is accessed by "SELECT * FROM ....". This is
implemented by fixing parse_relation.c.

> - 0008: Add aggregates support in IVM

This patch provides codes for aggregates support, specifically
for builtin count, sum, and avg.

When an IMMV containing an aggregate is created, it is checked if this
aggregate function is supported, and if it is ok, some hidden columns
are added to the target list.

When the IMMV is maintained, the aggregated value is updated as well as
related hidden columns. The way of update depends the type of aggregate
functions, and SET clause string is generated for each aggregate.

> - 0009: Add support for min/max aggregates for IVM

This patch adds min/max aggregates support.

This is separated from #0008 because min/max needs more complicated
work than count, sum, and avg.

If the view has min(x) or max(x) and the minimum or maximal value is
deleted from a table, we need to update the value to the new min/max
recalculated from the tables rather than incremental computation.
This is performed in recalc_and_set_values().

TIDs and keys of tuples that need re-calculation are returned as a
result of the query that deleted min/max values from the view using
RETURNING clause. The plan to recalculate and set the new min/max value
are stored and reused.

> - 0010: regression tests

This patch provides regression tests for IVM.

> - 0011: documentation

This patch provides documantation for IVM.

---------------------------------------------------------------------------------------
* Changes from the Previous Version (v27)

- Allow TRUNCATE on base tables

When a base table is truncated, the view content will be empty if the
view definition query does not contain an aggregate without a GROUP clause.
Therefore, such views can be truncated.

Aggregate views without a GROUP clause always have one row. Therefore,
if a base table is truncated, the view will not be empty and will contain
a row with NULL value (or 0 for count()). So, in this case, we refresh the
view instead of truncating it.

- Fix bugs reported by huyajun [1]

[1] https://www.postgresql.org/message-id/tencent_FCAF11BCA5003FD16BDDFDDA5D6A19587809%40qq.com

---------------------------------------------------------------------------------------
* Discussion

** Aggregate support

There were a few suggestions that general aggregate functions should be
supported [2][3], which may be possible by extending pg_aggregate catalog.
However, we decided to leave supporting general aggregates to the future work [4]
because it would need substantial works and make the patch more complex and
bigger.

There has been no opposite opinion on this. However, if we need more discussion
on the design of aggregate support, we can omit aggregate support for the first
release of IVM.

[2] https://www.postgresql.org/message-id/20191128140333.GA25947%40alvherre.pgsql
[3] https://www.postgresql.org/message-id/CAM-w4HOvDrL4ou6m%3D592zUiKGVzTcOpNj-d_cJqzL00fdsS5kg%40mail.gmail.com
[4] https://www.postgresql.org/message-id/20201016193034.9a4c44c79fc1eca7babe093e%40sraoss.co.jp

** Hidden columns

In order to support DISTINCT or aggregates, our implementation uses hidden columns.

Columns starting with "__ivm_" are hidden columns that doesn't appear when a
view is accessed by "SELECT * FROM ....". For this aim, parse_relation.c is
fixed. There was a proposal to enable hidden columns by adding a new flag to
pg_attribute [5], but this thread is no longer active, so we decided to check
the hidden column by its name [6].

[5] https://www.postgresql.org/message-id/flat/CAEepm%3D3ZHh%3Dp0nEEnVbs1Dig_UShPzHUcMNAqvDQUgYgcDo-pA%40mail.gmail.com
[6] https://www.postgresql.org/message-id/20201016193034.9a4c44c79fc1eca7babe093e%40sraoss.co.jp

** Concurrent Transactions

When the view definition has more than one table, we acquire an exclusive
lock before the view maintenance in order to avoid inconsistent results.
This behavior was explained in [7]. The lock was improved to use weaker lock
when the view has only one table based on a suggestion from Konstantin Knizhnik [8].
However, due to the implementation that uses ctid for identifying target tuples,
we still have to use an exclusive lock for DELETE and UPDATE.

[7] https://www.postgresql.org/message-id/20200909092752.c91758a1bec3479668e82643%40sraoss.co.jp
[8] https://www.postgresql.org/message-id/5663f5f0-48af-686c-bf3c-62d279567e2a%40postgrespro.ru

** Automatic Index Creation

When a view is created, a unique index is automatically created if
possible, that is, if the view definition query has a GROUP BY or
DISTINCT, or if the view contains all primary key attributes of
its base tables in the target list. It is necessary for efficient
view maintenance. This feature is based on a suggestion from
Konstantin Knizhnik [9].

[9] https://www.postgresql.org/message-id/89729da8-9042-7ea0-95af-e415df6da14d%40postgrespro.ru

** Trigger and Transition Tables

We implemented IVM based on triggers. This is because we want to use
transition tables to extract changes on base tables. Also, there are
other constraint that are using triggers in its implementation, like
foreign references. However, if we can use transition table like feature
without relying triggers, we don't have to insist to use triggers and we
might implement IVM in the executor directly as similar as declarative
partitioning.

** Feature to be Supported in the First Release

The current patch-set supports DISTINCT and aggregates for built-in count,
sum, avg, min and max. Do we need all these feature for the first IVM release?
Supporting DISTINCT and aggregates needs discussion on hidden columns, and
for supporting min/max we need to discuss on re-calculation method. Before
handling such relatively advanced feature, maybe, should we focus to design
and implement of the basic feature of IVM?

Any suggestion and discussion are welcomed!

Regards,
Yugo Nagata

--
Yugo NAGATA <nagata(at)sraoss(dot)co(dot)jp>