This topic describes how to write a commit timestamp for each insert and update
operation that you perform with Spanner. To use this feature, set the allow_commit_timestamp
option on a TIMESTAMP
column, then write
the timestamp as part of each transaction.
Overview
The commit timestamp, based on TrueTime
technology, is the time when a
transaction is committed in the database. The allow_commit_timestamp
column
option allows you to atomically store the commit timestamp into a column.
Using the commit timestamps stored in tables, you can determine the exact
ordering of mutations and build features like changelogs.
To insert commit timestamps in your database, complete the following steps:
-
Create a column with type
TIMESTAMPwith the column optionallow_commit_timestampset totruein the schema definition. For example:CREATE TABLE Performances ( ... LastUpdateTime TIMESTAMP NOT NULL OPTIONS ( allow_commit_timestamp = true ) ... ) PRIMARY KEY (...); -
If you are performing inserts or updates with DML, use the
PENDING_COMMIT_TIMESTAMPfunction to write the commit timestamp.If you are performing inserts or updates with mutations, use the placeholder string
spanner.commit_timestamp()on insertions or updates to your commit timestamp column. You can also use the commit timestamp constant provided by the client library. For example, this constant in the Java client isValue.COMMIT_TIMESTAMP.
When Spanner commits the transaction using these placeholders as column
values, the actual commit timestamp is written to the specified column (For
example: the LastUpdateTime
column). You could then use this column value
to create a history of updates to the table.
Commit timestamp values are not guaranteed to be unique. Transactions that write to non-overlapping sets of fields might have the same timestamp. Transactions that write to overlapping sets of fields have unique timestamps.
Spanner commit timestamps have microsecond granularity,
and they are converted to nanoseconds when stored in TIMESTAMP
columns.
Create and delete a commit timestamp column
Use the allow_commit_timestamp
column option to add and remove support for
commit timestamps:
- When creating a new table to specify that a column supports commit timestamps.
- When altering an existing table
:
- to add a new column supporting commit timestamps,
- to alter an existing
TIMESTAMPcolumn to support commit timestamps, - to alter an existing
TIMESTAMPcolumn to remove commit timestamp support
Keys and indexes
You can use a commit timestamp column as a primary key column or as a non-key
column. Primary keys can be defined as ASC
or DESC
.
-
ASC(default) - Ascending keys are ideal for answering queries from a specific time forward. -
DESC- Descending keys keep the latest rows at the top of the table. They provide quick access to the latest records.
The allow_commit_timestamp
option must be consistent across the primary
keys of parent and child tables. If the option is not consistent across
primary keys, Spanner returns an error. The only time the option
can be inconsistent is when you are creating or updating the schema.
Using commit timestamps under the following scenarios creates hotspots which reduce data performance:
-
Commit timestamp column as the first part of the primary key of a table:
CREATE TABLE Users ( LastAccess TIMESTAMP NOT NULL , UserId INT64 NOT NULL , ... ) PRIMARY KEY ( LastAccess , UserId ); -
The first part of the primary key of a secondary index:
CREATE INDEX UsersByLastAccess ON Users ( LastAccess )or
CREATE INDEX UsersByLastAccessAndName ON Users ( LastAccess , FirstName )
Hotspots reduce data performance, even with low write rates. There is no performance overhead if commit timestamps are enabled on non-key columns that are not indexed.
Create a commit timestamp column
The following DDL creates a table with a column that supports commit timestamps.
CREATE
TABLE
Performances
(
SingerId
INT64
NOT
NULL
,
VenueId
INT64
NOT
NULL
,
EventDate
Date
,
Revenue
INT64
,
LastUpdateTime
TIMESTAMP
NOT
NULL
OPTIONS
(
allow_commit_timestamp
=
true
)
)
PRIMARY
KEY
(
SingerId
,
VenueId
,
EventDate
),
INTERLEAVE
IN
PARENT
Singers
ON
DELETE
CASCADE
Adding the option changes the timestamp column as follows:
- You can use the
spanner.commit_timestamp()placeholder string (or a constant provided by the client library) for inserts and updates. - The column can only contain values in the past. For more information, see Providing your own value for the timestamp .
The option allow_commit_timestamp
is case sensitive.
Add a commit timestamp column to an existing table
To add a commit timestamp column to an existing table, use the ALTER TABLE
statement. For example to add a LastUpdateTime
column to the Performances
table, use the following statement:
ALTER
TABLE
Performances
ADD
COLUMN
LastUpdateTime
TIMESTAMP
NOT
NULL
OPTIONS
(
allow_commit_timestamp
=
true
)
Convert a timestamp column to a commit timestamp column
You can convert an existing timestamp column into a commit timestamp column, but doing so requires Spanner to validate that the existing timestamp values are in the past. For example:
ALTER
TABLE
Performances
ALTER
COLUMN
LastUpdateTime
SET
OPTIONS
(
allow_commit_timestamp
=
true
)
You cannot change the data type or NULL
annotation of a column in an ALTER TABLE
statement that includes SET OPTIONS
. For details, see Data Definition Language
.
Remove the commit timestamp option
If you want to remove commit timestamp support from a column, use the option allow_commit_timestamp=null
in an ALTER TABLE
statement. The commit timestamp behavior is removed, but the column is still
a timestamp. Changing the option does not alter any other characteristics of the
column, such as type or nullability ( NOT NULL
). For example:
ALTER
TABLE
Performances
ALTER
COLUMN
LastUpdateTime
SET
OPTIONS
(
allow_commit_timestamp
=
null
)
Write a commit timestamp using a DML statement
You use the PENDING_COMMIT_TIMESTAMP
function to write the commit
timestamp in a DML statement. Spanner selects the commit timestamp when the transaction
commits.
The following DML statement updates the LastUpdateTime
column in the Performances
table with the commit timestamp:
UPDATE
Performances
SET
LastUpdateTime
=
PENDING_COMMIT_TIMESTAMP
()
WHERE
SingerId
=
1
AND
VenueId
=
2
AND
EventDate
=
"2015-10-21"
The following code example uses the PENDING_COMMIT_TIMESTAMP
function to write the commit timestamp in the LastUpdateTime
column.
C++
C#
Go
Java
Node.js
PHP
Python
Ruby
Ruby
Commit timestamps can only be written to columns annotated with the allow_commit_timestamp=true
option.
If you have mutations on rows in multiple tables, you must specify spanner.commit_timestamp()
(or the client library
constant) for the commit timestamp column in each table.
Query a commit timestamp column
The following example queries the commit timestamp column of the table.
C++
C#
Go
Java
Node.js
PHP
Python
Ruby
Provide your own value for the commit timestamp column
You can provide your own value for the commit timestamp column, instead of
passing spanner.commit_timestamp()
(or client library constant) as the
column value. The value must be a timestamp in the past. This restriction
ensures that writing timestamps is an inexpensive and fast operation. The server
returns a FailedPrecondition
error if a future timestamp is specified.
Create a changelog
Suppose that you want to create a changelog of every mutation that happens to a table and then use that changelog for auditing. An example would be a table that stores the history of changes to word processing documents. The commit timestamp makes creating the changelog easier, because the timestamps can enforce ordering of the changelog entries. You could build a changelog that stores the history of changes to a given document using a schema like the following example:
CREATE
TABLE
Documents
(
UserId
INT64
NOT
NULL
,
DocumentId
INT64
NOT
NULL
,
Contents
STRING
(
MAX
)
NOT
NULL
,
)
PRIMARY
KEY
(
UserId
,
DocumentId
);
CREATE
TABLE
DocumentHistory
(
UserId
INT64
NOT
NULL
,
DocumentId
INT64
NOT
NULL
,
Ts
TIMESTAMP
NOT
NULL
OPTIONS
(
allow_commit_timestamp
=
true
),
Delta
STRING
(
MAX
),
)
PRIMARY
KEY
(
UserId
,
DocumentId
,
Ts
),
INTERLEAVE
IN
PARENT
Documents
ON
DELETE
NO
ACTION
;
To create a changelog, insert a new row in DocumentHistory
in the same
transaction in which you insert or update a row in Document
. In the insertion
of the new row in DocumentHistory
, use the placeholder spanner.commit_timestamp()
(or client library constant) to tell
Spanner to write the commit timestamp into column Ts
. Interleaving
the DocumentsHistory
table with the Documents
table will allow for data
locality and more efficient inserts and updates. However, it also adds the
constraint that the parent and child rows
must be deleted together. To keep the rows in DocumentHistory
after rows
in Documents
are deleted, do not interleave the tables.
Optimize recent-data queries with commit timestamps
Commit timestamps enable a Spanner optimization that can reduce query I/O when retrieving data written after a particular time.
To activate this optimization, a query's WHERE
clause must include a
comparison between a table's commit timestamp column and a specific time
that you provide, with the following attributes:
-
Provide the specific time as a constant expression : a literal, a parameter, or a function whose own arguments evaluate to constants.
-
Compare whether the commit timestamp is more recent than the given time, through either the
>or>=operators. -
Optionally, add further restrictions to the
WHEREclause withAND. Extending the clause withORdisqualifies the query from this optimization.
For example, consider the following Performances
table, which includes
a commit timestamp column:
CREATE
TABLE
Performances
(
SingerId
INT64
NOT
NULL
,
VenueId
INT64
NOT
NULL
,
EventDate
DATE
,
Revenue
INT64
,
LastUpdateTime
TIMESTAMP
NOT
NULL
OPTIONS
(
allow_commit_timestamp
=
true
)
)
PRIMARY
KEY
(
SingerId
,
VenueId
,
EventDate
);
This query benefits from the commit-timestamp optimization described earlier, because it has a greater-than-or-equal-to comparison between the table's commit timestamp column and a constant expression—in this case, a literal:
SELECT
*
FROM
Performances
WHERE
LastUpdateTime
> =
"2022-05-01"
;
The following query also qualifies for the optimization, since it has a greater-than comparison between the commit timestamp and a function whose arguments all evaluate to constants during the query's execution:
SELECT
*
FROM
Performances
WHERE
LastUpdateTime
>
TIMESTAMP_SUB
(
CURRENT_TIMESTAMP
(),
INTERVAL
30
DAY
);
What's next
- Use commit timestamps to create a change log with Go .
