Name

sqitchtutorial-clickhouse - A tutorial introduction to Sqitch change management on ClickHouse

Synopsis

sqitch *

Description

This tutorial explains how to create a sqitch-enabled ClickHouse project, use a VCS for deployment planning, and work with other developers to make sure changes remain in sync and in the proper order.

We’ll start by creating a new project from scratch, a fictional antisocial networking site called Flipr. All examples use Git as the VCS and ClickHouse as the storage engine, but for the most part you can substitute other VCSes and database engines in the examples as appropriate.

If you’d like to manage a database with a different database engine, see one of the following other tutorials:

• PostgreSQL, Yugabyte, and CockroachDB Tutorial
• SQLite Tutorial
• Oracle Tutorial
• MySQL and MariaDB Tutorial
• Firebird Tutorial
• Vertica Tutorial
• Exasol Tutorial
• Snowflake Tutorial
• ClickHouse Tutorial

Connection Configuration

Sqitch requires ClickHouse v25.8 or higher for its support for lightweight updates and deletes. If you’ve upgraded an instance from an earlier version, be sure to SET compatibility = '25.8' or higher.

Sqitch requires ODBC to connect to the ClickHouse database. As such, you’ll need to make sure that the ClickHouse ODBC driver is installed and properly configured. At its simplest, on Unix-like systems, name the driver “ClickHouse” by adding this entry to odbcinst.ini (usually found in /etc, /usr/etc, or /usr/local/etc):

[ClickHouse]
Description = Unicode ODBC Driver for ClickHouse
Driver      = /opt/clickhouse/lib/libclickhouseodbcw.so

Note the use of the Unicode rather than the ANSI driver. Also, you might need to adjust the path depending on the version of the ODBC driver, and where you installed it. Consult the documentation for your ODBC driver manager (e.g., UnixODBC or iODBC) for details.

With this configured, Sqitch target URIs can specify the driver name like so:

db:clickhouse://user:pass@host:port/dbname?Driver=ClickHouse

Ports

In addition to the ODBC connection Sqitch uses to maintain its registry tables, it depends on the ClickHouse Client|https://clickhouse.com/docs/interfaces/cli to run change scripts. As such, Sqitch requires on both the HTTP and Native ports to connect.\ The port provided as part of a URI will determines the HTTP port used by ODBC. Therefore, if your ClickHouse server uses a non-standard Native port (other than 9004 or 9440 for encrypted connections), you’ll need to specify the native port via the NativePort query parameter:

db:clickhouse://user:pass@host:port/dbname?Driver=ClickHouse?NativePort=9999

Starting a New Project

Usually the first thing to do when starting a new project is to create a source code repository. So let’s do that with Git:

> mkdir flipr
> cd flipr
> git init .
Initialized empty Git repository in /flipr/.git/
> touch README.md
> git add .
> git commit -am 'Initialize project, add README.'
[main (root-commit) 046b8f5] Initialize project, add README.
 1 file changed, 38 insertions(+)
 create mode 100644 README.md

If you’re a Git user and want to follow along the history, the repository used in these examples is on GitHub.

Now that we have a repository, let’s get started with Sqitch. Every Sqitch project must have a name associated with it, and, optionally, a unique URI. We recommend including the URI, as it increases the uniqueness of object identifiers internally, and will prevent the deployment of a different project with the same name. So let’s specify one when we initialize Sqitch:

> sqitch init flipr --uri https://github.com/sqitchers/sqitch-clickhouse-intro/ --engine clickhouse
Created sqitch.conf
Created sqitch.plan
Created deploy/
Created revert/
Created verify/

Let’s have a look at sqitch.conf:

> cat sqitch.conf
[core]
        engine = clickhouse
        # plan_file = sqitch.plan
        # top_dir = .
# [engine "clickhouse"]
        # target = db:clickhouse:
        # registry = sqitch
        # client = clickhouse-client

Good, it picked up on the fact that we’re creating changes for the ClickHouse engine, thanks to the --engine clickhouse option, and saved it to the file. Furthermore, it wrote a commented-out [engine "clickhouse"] section with ClickHouse engine-specific settings commented out and ready to be edited as appropriate.

By default, Sqitch will read sqitch.conf in the current directory for settings. But it will also read ~/.sqitch/sqitch.conf for user-specific settings. Since ClickHouse’s clickhouse client is not in the path on my system, let’s go ahead an tell it where to find the client on our computer (don’t bother if you’re using the Docker image because it uses the client inside the container, not on your host machine):

> sqitch config --user engine.clickhouse.client /opt/bin/clickhouse

And let’s also tell it who we are, since this data will be used in all of our projects:

> sqitch config --user user.name 'Marge N. O’Vera'
> sqitch config --user user.email 'marge@example.com'

Have a look at ~/.sqitch/sqitch.conf and you’ll see this:

> cat ~/.sqitch/sqitch.conf
[engine "clickhouse"]
        client = /opt/bin/clickhouse
[user]
        name = Marge N. O’Vera
        email = marge@example.com

Which means that Sqitch should be able to find clickhouse for any project, and that it will always properly identify us when planning and committing changes.

Back to the repository. Have a look at the plan file, sqitch.plan:

> cat sqitch.plan
%syntax-version=1.0.0
%project=flipr
%uri=https://github.com/sqitchers/sqitch-clickhouse-intro/

Note that it contains the name and URI of the app we’re building. Sqitch uses this data to manage cross-project dependencies. The %syntax-version pragma is always set by Sqitch, so that it always knows how to parse the plan, even if the format changes in the future.

Let’s commit these changes and start creating the database changes.

> git add .
> git commit -am 'Initialize Sqitch configuration.'
[main d32b8ae] Initialize Sqitch configuration.
 2 files changed, 12 insertions(+)
 create mode 100644 sqitch.conf
 create mode 100644 sqitch.plan

Our First Change

Let’s create a table. Our app will need users, of course, so we’ll create a table for them. Run this command:

> sqitch add users -n 'Creates table to track our users.'
Created deploy/users.sql
Created revert/users.sql
Created verify/users.sql
Added "users" to sqitch.plan

The add command adds a database change to the plan and writes deploy, revert, and verify scripts that represent the change. Now we edit these files. The deploy script’s job is to create the table. By default, the deploy/users.sql file looks like this:

-- Deploy flipr:users to clickhouse

-- XXX Add DDLs here.

COMMIT;

What we want to do is to replace the XXX comment with the CREATE TABLE statement, like so:

-- Deploy flipr:users to clickhouse

CREATE TABLE users (
    nickname   VARCHAR(50)  PRIMARY KEY,
    password   VARCHAR(512) NOT NULL,
    fullname   VARCHAR(512) NOT NULL,
    mastodon   VARCHAR(512) NOT NULL,
    created_at DATETIME64(6, 'UTC') NOT NULL DEFAULT now64(6, 'UTC')
) ENGINE = MergeTree;

The revert script’s job is to precisely revert the change to the deploy script, so we edit this to revert/users.sql to look like this:

-- Revert flipr:users from clickhouse

DROP TABLE users;

Now we can try deploying this change. First, we need to create a database to deploy to. Assuming that ClickHouse is running on localhost using the default user (as is provided by the ClickHouse Docker Image):

> clickhouse client --query 'CREATE DATABASE flipr_test'

Now we tell Sqitch where to send the change via a database URI, again assuming the default user & database and an ODBC driver named ClickHouse (see “Connection Configuration” for details):

> sqitch deploy 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Adding registry tables to db:clickhouse://default@localhost/sqitch?Driver=ClickHouse
Deploying changes to db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
  + users .. ok

First Sqitch created the registry database and tables used to track database changes. The registry database is separate from the database to which the users change was deployed; by default, its name is sqitch, and will be used to manage all projects on a single ClickHouse server. Ideally, only Sqitch data will be stored in this database, so it probably makes the most sense to create a superuser named sqitch or something similar and use it to deploy changes.

If you’d like it to use a different database as the registry database, use sqitch engine add clickhouse $name to configure it (or via the target command; more below). This will be useful if you don’t want to use the same registry database to manage multiple databases on the same server.

Next, Sqitch deploys changes to the target database, which we specified on the command-line. We only have one change so far; the + reinforces the idea that the change is being added to the database.

With this change deployed, if you connect to the database, you’ll be able to see the user:

> clickhouse client -d flipr_test -q 'show tables'
users

Trust, But Verify

But that’s too much work. Do you really want to do something like that after every deploy?

Here’s where the verify script comes in. Its job is to test that the deploy did was it was supposed to. It should do so without regard to any data that might be in the database, and should throw an error if the deploy was not successful. The easiest way to do that with a table is to simply SELECT from it. Put this query into verify/users.sql:

SELECT nickname, password, fullname, mastodon
  FROM users
 WHERE false;

Now you can run the verify script with the verify command:

> sqitch verify 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Verifying db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
  * users .. ok
Verify successful

Looks good! If you want to make sure that the verify script correctly dies if the table doesn’t exist, temporarily change the table name in the script to something that doesn’t exist, something like:

SELECT nickname, password, fullname, mastodon, created_at
  FROM users_nonesuch
 WHERE false;

Then verify again:

> sqitch verify 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Verifying db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
  * users .. Received exception from server (version 25.8.2):
Code: 60. DB::Exception: Received from clickhouse:9000. DB::Exception: Unknown table expression identifier 'users_nonesuch' in scope SELECT nickname, password, fullname, mastodon FROM users_nonesuch WHERE false. (UNKNOWN_TABLE)
(query: -- Verify flipr:users on clickhouse

SELECT nickname, password, fullname, mastodon
  FROM users_nonesuch
 WHERE false;)
# Verify script "verify/users.sql" failed.
not ok

Verify Summary Report
---------------------
Changes: 1
Errors:  1
Verify failed

ClickHouse is kind enough to tell us what the problem is. Don’t forget to change the table name back before continuing!

Status, Revert, Log, Repeat

For purely informational purposes, we can always see how a deployment was recorded via the status command, which reads the registry tables from the database:

> sqitch status 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
# On database db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
# Project:  flipr
# Change:   02916d6d5dca9faef8f2bb5822c53de8b7ae16bc
# Name:     users
# Deployed: 2025-09-19 14:32:08 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

Let’s make sure that we can revert the change:

> sqitch revert 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Revert all changes from db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse? [Yes]
  - users .. ok

The revert command first prompts to make sure that we really do want to revert. This is to prevent unnecessary accidents. You can pass the -y option to disable the prompt. Also, notice the - before the change name in the output, which reinforces that the change is being removed from the database. And now the schema should be gone:

> clickhouse client -d flipr_test -q 'show tables'

And the status message should reflect as much:

> sqitch status 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
# On database db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
No changes deployed

Of course, since nothing is deployed, the verify command has nothing to verify:

> sqitch verify 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Verifying db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
No changes deployed

However, we still have a record that the change happened, visible via the log command:

> sqitch log 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Revert 02916d6d5dca9faef8f2bb5822c53de8b7ae16bc
Name:      users
Committer: Marge N. O’Vera <marge@example.com>
Date:      2025-09-19 14:33:23 -0400

    Creates table to track our users.

Deploy 02916d6d5dca9faef8f2bb5822c53de8b7ae16bc
Name:      users
Committer: Marge N. O’Vera <marge@example.com>
Date:      2025-09-19 14:32:08 -0400

    Creates table to track our users.

Note that the actions we took are shown in reverse chronological order, with the revert first and then the deploy.

Cool. Now let’s commit it.

> git add .
> git commit -m 'Add users table.'
[main 414672a] Add users table.
 4 files changed, 18 insertions(+)
 create mode 100644 deploy/users.sql
 create mode 100644 revert/users.sql
 create mode 100644 verify/users.sql

And then deploy again. This time, let’s use the --verify option, so that the verify script is applied when the change is deployed:

> sqitch deploy --verify 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
Deploying changes to db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
  + users .. ok

And now the schema should be back:

> clickhouse client -d flipr_test -q 'show tables'
users

When we look at the status, the deployment will be there:

> sqitch status 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'
# On database db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse
# Project:  flipr
# Change:   02916d6d5dca9faef8f2bb5822c53de8b7ae16bc
# Name:     users
# Deployed: 2025-09-19 14:34:21 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

On Target

I’m getting a little tired of always having to type db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse, aren’t you? This database connection URI tells Sqitch how to connect to the deployment target, but we don’t have to keep using the URI. We can name the target:

> sqitch target add flipr_test 'db:clickhouse://default@localhost/flipr_test?Driver=ClickHouse'

The target command, inspired by git-remote, allows management of one or more named deployment targets. We’ve just added a target named flipr_test, which means we can use the string flipr_test for the target, rather than the URI. But since we’re doing so much testing, we can also tell Sqitch to deploy to the flipr_test target by default:

> sqitch engine add clickhouse flipr_test

Now we can omit the target argument altogether, unless we need to deploy to another database. Which we will, eventually, but at least our examples will be simpler from here on in, e.g.:

> sqitch status
# On database flipr_test
# Project:  flipr
# Change:   02916d6d5dca9faef8f2bb5822c53de8b7ae16bc
# Name:     users
# Deployed: 2025-09-19 14:34:21 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

Yay, that allows things to be a little more concise. Let’s also make sure that changes are verified after deploying them:

> sqitch config --bool deploy.verify true
> sqitch config --bool rebase.verify true

We’ll see the rebase command a bit later. In the meantime, let’s commit the new configuration and and make some more changes!

> git commit -am 'Set default deployment target and always verify.'
[main ad3a3b0] Set default deployment target and always verify.
 1 file changed, 8 insertions(+)

Deploy with Dependency

Let’s add another change. Our app will need to store status messages from users. Let’s call them – and the table to store them – “flips”. First, add the new change:

> sqitch add flips --requires users -n 'Adds table for storing flips.'
Created deploy/flips.sql
Created revert/flips.sql
Created verify/flips.sql
Added "flips [users]" to sqitch.plan

Note that we’re requiring the users change as a dependency of the new flips change. Although that change has already been added to the plan and therefore should always be applied before the flips change, it’s a good idea to be explicit about dependencies.

Now edit the scripts. When you’re done, deploy/flips.sql should look like this:

-- Deploy flipr:flips to clickhouse
-- requires: users

CREATE TABLE flips (
    id         INTEGER       NOT NULL PRIMARY KEY,
    nickname   VARCHAR(50)   NOT NULL,
    body       VARCHAR(512)  DEFAULT '' NOT NULL,
    created_at DATETIME64(6, 'UTC') NOT NULL DEFAULT now64(6, 'UTC'),
    CONSTRAINT body_length_check CHECK ( char_length(body) <= 180 ),
) ENGINE = MergeTree;

A couple things to notice here. On the second line, the dependence on the users change has been listed. This doesn’t do anything, but the default deploy template lists it here for your reference while editing the file. Useful, right?

The users.nickname column implicitly references the users table. This is why we need to require the users change, even though ClickHouse itself does not enforce such relationships.

Now for the verify script. Again, all we need to do is SELECT from the table. I recommend selecting each column by name, too, to be sure that no column is missing. Here’s the verify/flips.sql:

-- Verify flipr:flips on clickhouse

SELECT id, nickname, body, created_at
  FROM flips
 WHERE false;

Now for the revert script: all we have to do is drop the table. Add this to revert/flips.sql:

-- Revert flipr:flips from firebird

DROP TABLE flips;

Couldn’t be much simpler, right? Let’s deploy this bad boy:

> sqitch deploy
Deploying changes to flipr_test
  + flips .. ok

We know, since verification is enabled, that the table must have been created. But for the purposes of visibility, let’s have a quick look:

> clickhouse client -d flipr_test -q 'show tables'
flips
users

We can also verify all currently deployed changes with the verify command:

> sqitch verify
Verifying flipr_test
  * users .. ok
  * flips .. ok
Verify successful

Now have a look at the status:

> sqitch status
# On database flipr_test
# Project:  flipr
# Change:   0756d81da8a858a991dca5ffd167fea2de8e98df
# Name:     flips
# Deployed: 2025-09-19 14:36:09 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

Success! Let’s make sure we can revert the change, as well:

> sqitch revert --to @HEAD^ -y
Reverting changes to users from flipr_test
  - flips .. ok

Note that we’ve used the --to option to specify the change to revert to. And what do we revert to? The symbolic tag @HEAD, when passed to revert, always refers to the last change deployed to the database. (For other commands, it refers to the last change in the plan.) Appending the caret (^) tells Sqitch to select the change prior to the last deployed change. So we revert to users, the penultimate change. The other potentially useful symbolic tag is @ROOT, which refers to the first change deployed to the database (or in the plan, depending on the command).

Back to the database. The flips table should be gone but the users table should still be around:

> clickhouse client -d flipr_test -q 'show tables'
users

The status command politely informs us that we have undeployed changes:

> sqitch status
# On database flipr_test
# Project:  flipr
# Change:   02916d6d5dca9faef8f2bb5822c53de8b7ae16bc
# Name:     users
# Deployed: 2025-09-19 14:34:21 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Undeployed change:
  * flips

As does the verify command:

> sqitch verify
Verifying flipr_test
  * users .. ok
Undeployed change:
  * flips
Verify successful

Note that the verify is successful, because all currently-deployed changes are verified. The list of undeployed changes (just “flips” here) reminds us about the current state.

Okay, let’s commit and deploy again:

> git add .
> git commit -am 'Add flips table.'
[main 347f867] Add flips table.
 4 files changed, 19 insertions(+)
 create mode 100644 deploy/flips.sql
 create mode 100644 revert/flips.sql
 create mode 100644 verify/flips.sql
> sqitch deploy
Deploying changes to flipr_test
  + flips .. ok

Looks good. Check the status:

> sqitch status
# On database flipr_test
# Project:  flipr
# Change:   0756d81da8a858a991dca5ffd167fea2de8e98df
# Name:     flips
# Deployed: 2025-09-19 14:38:00 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

View to a Thrill

One more thing to add before we are ready to ship a first beta release. Let’s create a view that lists user names with their flips.

> sqitch add userflips --requires users --requires flips \
  -n 'Creates the userflips view.'
Created deploy/userflips.sql
Created revert/userflips.sql
Created verify/userflips.sql
Added "userflips [users flips]" to sqitch.plan

Now add this SQL to deploy/userflips.sql:

CREATE OR REPLACE VIEW userflips AS
SELECT f.id, u.nickname, u.fullname, f.body, f.created_at AS created_at
  FROM users u
  JOIN flips f ON u.nickname = f.nickname;

Add this SQL to verify/userflips.sql

SELECT id, nickname, fullname, body, created_at
  FROM userflips
 WHERE false;

And add the DROP VIEW statement to revert/userflips.sql:

DROP VIEW userflips;

Now Try it out!

> sqitch deploy
Deploying changes to flipr_test
  + userflips .. ok
> sqitch revert -y
Reverting all changes from flipr_test
  - userflips .. ok
  - flips ...... ok
  - users ...... ok
> sqitch deploy
Deploying changes to flipr_test
  + users ...... ok
  + flips ...... ok
  + userflips .. ok

Looks good! Commit it.

> git add .
> git commit -m 'Add the userflips view.'
[main f21de26] Add the userflips view.
 4 files changed, 17 insertions(+)
 create mode 100644 deploy/userflips.sql
 create mode 100644 revert/userflips.sql
 create mode 100644 verify/userflips.sql

Ship It!

Now we’re ready for the first development release of our app. Let’s call it 1.0.0-dev1 Since we want to have it go out with deployments tied to the release, let’s tag it:

> sqitch tag v1.0.0-dev1 -n 'Tag v1.0.0-dev1.'
Tagged "userflips" with @v1.0.0-dev1
> git commit -am 'Tag the database with v1.0.0-dev1.'
[main a821008] Tag the database with v1.0.0-dev1.
 1 files changed, 1 insertions(+), 0 deletions(-)
> git tag v1.0.0-dev1 -am 'Tag v1.0.0-dev1'

Now let’s bundle everything up for release:

> sqitch bundle
Bundling into bundle
Writing config
Writing plan
Writing scripts
  + users
  + flips
  + userflips @v1.0.0-dev1

Now we can package the bundle directory and distribute it. When it gets installed somewhere, users can use Sqitch to deploy to the database. We ought to try deploying it, but first we’ll need to revert our existing databases, as a single Sqitch project cannot be deployed to two databases on the same server unless it uses a different registry database. Fortunately, it’s easy to build the database again, so let’s just revert it.

> sqitch revert -y
Reverting all changes from flipr_test
  - userflips @v1.0.0-dev1 .. ok
  - flips ................... ok
  - users ................... ok

Now we can try deploying the bundle:

> cd bundle
> clickhouse client --query 'CREATE DATABASE flipr_dev'
> sqitch deploy 'db:clickhouse://default@localhost/flipr_dev?Driver=ClickHouse'
Deploying changes to db:clickhouse://default@localhost/flipr_dev?Driver=ClickHouse
  + users ................... ok
  + flips ................... ok
  + userflips @v1.0.0-dev1 .. ok

Great, all four changes were deployed and userflips was tagged with @v1.0.0-dev1. Let’s have a look at the status:

> sqitch status 'db:clickhouse://default@localhost/flipr_dev?Driver=ClickHouse'
# On database db:clickhouse://default@localhost/flipr_dev?Driver=ClickHouse
# Project:  flipr
# Change:   afdfa60d0fd2b7910601648667a23546fe0f7f43
# Name:     userflips
# Tag:      @v1.0.0-dev1
# Deployed: 2025-09-19 14:40:20 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

Looks good, eh? Go ahead and revert it:

> sqitch revert -y 'db:clickhouse://default@localhost/flipr_dev?Driver=ClickHouse'
Revert all changes from db:clickhouse://default@localhost/flipr_dev?Driver=ClickHouse? [Yes]
  - userflips @v1.0.0-dev1 .. ok
  - flips ................... ok
  - users ................... ok

Now package it up and ship it!

> cd ..
> mv bundle flipr-v1.0.0-dev1
> tar -czf flipr-v1.0.0-dev1.tgz flipr-v1.0.0-dev1

Making a Hash of Things

Now that we’ve got the basics of the app done, let’s add a feature. Gotta track the hashtags associated with flips, right? Let’s add a table for them. Since other folks are working on other tasks in the repository, we’ll work on a branch, so we can all stay out of each other’s way. So let’s branch:

> git checkout -b hashtags
Switched to a new branch 'hashtags'

Now we can add a new change to create a table for hashtags.

> sqitch add hashtags --requires flips -n 'Adds table for storing hashtags.'
Created deploy/hashtags.sql
Created revert/hashtags.sql
Created verify/hashtags.sql
Added "hashtags [flips]" to sqitch.plan

You know the drill by now. Add this to deploy/hashtags.sql

CREATE TABLE hashtags (
    flip_id   INTEGER      NOT NULL,
    hashtag   VARCHAR(512) NOT NULL,
    CONSTRAINT hashtag CHECK(char_length(hashtag) > 0)
)  ENGINE = MergeTree
   PRIMARY KEY (flip_id, hashtag);

Again, select from the table in verify/hashtags.sql:

SELECT flip_id, hashtag FROM hashtags WHERE false;

And drop it in revert/hashtags.sql

DROP TABLE hashtags;

And give it a whirl:

> sqitch deploy
Deploying changes to flipr_test
  + users ................... ok
  + flips ................... ok
  + userflips @v1.0.0-dev1 .. ok
  + hashtags ................ ok

Look good?

> sqitch status --show-tags
# On database flipr_test
# Project:  flipr
# Change:   dca0003c3d411c71a672172acab78e4a42dfaa8d
# Name:     hashtags
# Deployed: 2025-09-19 14:48:13 -0400
# By:       Marge N. O’Vera <marge@example.com>
# 
# Tag:
#   @v1.0.0-dev1 - 2025-09-19 14:48:13 -0400 - Marge N. O’Vera <marge@example.com>
# 
Nothing to deploy (up-to-date)

Note the use of --show-tags to show all the deployed tags. Now make it so:

> rm -rf flipr-v1.0.0-dev1*
> git add .
> git commit -am 'Add hashtags table.'
[hashtags 1359929] Add hashtags table.
 4 files changed, 17 insertions(+)
 create mode 100644 deploy/hashtags.sql
 create mode 100644 revert/hashtags.sql
 create mode 100644 verify/hashtags.sql

Good, we’ve finished this feature. Time to merge back into main.

Emergency

Let’s do it:

> git checkout main
Switched to branch 'main'
> git pull
Updating a821008..b800878
Fast-forward
 deploy/lists.sql | 10 ++++++++++
 revert/lists.sql |  3 +++
 sqitch.plan      |  2 ++
 verify/lists.sql |  5 +++++
 4 files changed, 20 insertions(+)
 create mode 100644 deploy/lists.sql
 create mode 100644 revert/lists.sql
 create mode 100644 verify/lists.sql

Hrm, that’s interesting. Looks like someone made some changes to main. They added list support. Well, let’s see what happens when we merge our changes.

> git merge --no-ff hashtags
Auto-merging sqitch.plan
CONFLICT (content): Merge conflict in sqitch.plan
Automatic merge failed; fix conflicts and then commit the result.

Oh, a conflict in sqitch.plan. Not too surprising, since both the merged lists branch and our hashtags branch added changes to the plan. Let’s try a different approach.

The truth is, we got lazy. Those changes when we pulled main from the origin should have raised a red flag. It’s considered a bad practice not to look at what’s changed in main before merging in a branch. What one should do is either:

• Rebase the hashtags branch from main before merging. This “rewinds” the branch changes, pulls from main, and then replays the changes back on top of the pulled changes.
• Create a patch and apply that to main. This is the sort of thing you might have to do if you’re sending changes to another user, especially if the VCS is not Git.

So let’s restore things to how they were at main:

> git merge --abort

That throws out our botched merge. Now let’s go back to our branch and rebase it on main:

> git checkout hashtags
Switched to branch 'hashtags'
> git rebase main
Auto-merging sqitch.plan
CONFLICT (content): Merge conflict in sqitch.plan
error: could not apply 1359929... Add hashtags table.
hint: Resolve all conflicts manually, mark them as resolved with
hint: "git add/rm <conflicted_files>", then run "git rebase --continue".
hint: You can instead skip this commit: run "git rebase --skip".
hint: To abort and get back to the state before "git rebase", run "git rebase --abort".
hint: Disable this message with "git config set advice.mergeConflict false"
Could not apply 1359929... # Add hashtags table.

Oy, that’s kind of a pain. It seems like no matter what we do, we’ll need to resolve conflicts in that file. Except in Git. Fortunately for us, we can tell Git to resolve conflicts in sqitch.plan differently. Because we only ever append lines to the file, we can have it use the “union” merge driver, which, according to its docs:

Run 3-way file level merge for text files, but take lines from both versions, instead of leaving conflict markers. This tends to leave the added lines in the resulting file in random order and the user should verify the result. Do not use this if you do not understand the implications.

This has the effect of appending lines from all the merging files, which is exactly what we need. So let’s give it a try. First, back out the botched rebase:

> git rebase --abort

Now add the union merge driver to .gitattributes for sqitch.plan and rebase again:

> echo sqitch.plan merge=union > .gitattributes
> git rebase main
Successfully rebased and updated refs/heads/hashtags.

Ah, that looks a bit better. Let’s have a look at the plan:

> cat sqitch.plan
%syntax-version=1.0.0
%project=flipr
%uri=https://github.com/sqitchers/sqitch-clickhouse-intro/

users 2025-09-19T18:29:21Z Marge N. O’Vera <marge@example.com> # Creates table to track our users.
flips [users] 2025-09-19T18:35:31Z Marge N. O’Vera <marge@example.com> # Adds table for storing flips.
userflips [users flips] 2025-09-19T18:38:26Z Marge N. O’Vera <marge@example.com> # Creates the userflips view.
@v1.0.0-dev1 2025-09-19T18:39:48Z Marge N. O’Vera <marge@example.com> # Tag v1.0.0-dev1.

lists [flips] 2025-09-19T18:41:17Z An Oth R. Developer <another@example.com> # Adds table for storing lists.
hashtags [flips] 2025-09-19T18:47:02Z Marge N. O’Vera <marge@example.com> # Adds table for storing hashtags.

Note that it has appended the changes from the merged “lists” branch, and then merged the changes from our “hashtags” branch. Test it to make sure it works as expected:

> sqitch rebase -y
Reverting all changes from flipr_test
  - hashtags ................ ok
  - userflips @v1.0.0-dev1 .. ok
  - flips ................... ok
  - users ................... ok
Deploying changes to flipr_test
  + users ................... ok
  + flips ................... ok
  + userflips @v1.0.0-dev1 .. ok
  + lists ................... ok
  + hashtags ................ ok

Note the use of rebase, which combines a revert and a deploy into a single command. Handy, right? It correctly reverted our changes, and then deployed them all again in the proper order. So let’s commit .gitattributes; seems worthwhile to keep that change:

> git add .
> git commit -m 'Add `.gitattributes` with union merge for `sqitch.plan`.'
[hashtags 862ea7a] Add `.gitattributes` with union merge for `sqitch.plan`.
 1 file changed, 1 insertion(+)
 create mode 100644 .gitattributes

Merges Mastered

And now, finally, we can merge into main:

> git checkout main
Switched to branch 'main'
> git merge --no-ff hashtags -m "Merge branch 'hashtags'"
Merge made by the 'ort' strategy.
 .gitattributes      | 1 +
 deploy/hashtags.sql | 9 +++++++++
 revert/hashtags.sql | 3 +++
 sqitch.plan         | 1 +
 verify/hashtags.sql | 3 +++
 5 files changed, 17 insertions(+)
 create mode 100644 .gitattributes
 create mode 100644 deploy/hashtags.sql
 create mode 100644 revert/hashtags.sql
 create mode 100644 verify/hashtags.sql

And double-check our work:

> cat sqitch.plan
%syntax-version=1.0.0
%project=flipr
%uri=https://github.com/sqitchers/sqitch-clickhouse-intro/

users 2025-09-19T18:29:21Z Marge N. O’Vera <marge@example.com> # Creates table to track our users.
flips [users] 2025-09-19T18:35:31Z Marge N. O’Vera <marge@example.com> # Adds table for storing flips.
userflips [users flips] 2025-09-19T18:38:26Z Marge N. O’Vera <marge@example.com> # Creates the userflips view.
@v1.0.0-dev1 2025-09-19T18:39:48Z Marge N. O’Vera <marge@example.com> # Tag v1.0.0-dev1.

lists [flips] 2025-09-19T18:41:17Z An Oth R. Developer <another@example.com> # Adds table for storing lists.
hashtags [flips] 2025-09-19T18:47:02Z Marge N. O’Vera <marge@example.com> # Adds table for storing hashtags.

Much much better, a nice clean main now. And because it is now identical to the “hashtags” branch, we can just carry on. Go ahead and tag it, bundle, and release:

> sqitch tag v1.0.0-dev2 -n 'Tag v1.0.0-dev2.'
Tagged "hashtags" with @v1.0.0-dev2
> git commit -am 'Tag the database with v1.0.0-dev2.'
[main 5b56772] Tag the database with v1.0.0-dev2.
 1 file changed, 1 insertion(+)
> git tag v1.0.0-dev2 -am 'Tag v1.0.0-dev2'
> sqitch bundle --dest-dir flipr-1.0.0-dev2
Bundling into flipr-1.0.0-dev2
Writing config
Writing plan
Writing scripts
  + users
  + flips
  + userflips @v1.0.0-dev1
  + lists
  + hashtags @v1.0.0-dev2

Note the use of the --dest-dir option to sqitch bundle. Just a nicer way to create the top-level directory name so we don’t have to rename it from bundle.

In Place Changes

Well, some folks have been testing the 1.0.0-dev2 release and have demanded that Mastodon user links be added to Flipr pages. Why anyone would want to include social network links in an anti-social networking app is beyond us programmers, but we’re just the plumbers, right? Gotta go with what Marketing demands. The upshot is that we need to update the userflips view, which is used for the feature in question, to include the Mastodon user names.

Normally, modifying views in database changes is a PITA. You have to make changes like these:

1. Copy deploy/userflips.sql to deploy/userflips_mastodon.sql.
2. Edit deploy/userflips_mastodon.sql to re-create the view with the new mastodon column added to the view.
3. Copy deploy/userflips.sql to revert/userflips_mastodon.sql. Yes, copy the original change script to the new revert change.
4. Add a DROP VIEW statement to revert/userflips_mastodon.sql.
5. Copy verify/userflips.sql to verify/userflips_mastodon.sql.
6. Modify verify/userflips_mastodon.sql to include a check for the mastodon column.
7. Test the changes to make sure you can deploy and revert the userflips_mastodon change.

But you can have Sqitch do most of the work for you. The only requirement is that a tag appear between the two instances of a change we want to modify. In general, you’re going to make a change like this after a release, which you’ve tagged anyway, right? Well we have, with @v1.0.0-dev2 added in the previous section. With that, we can let Sqitch do most of the hard work for us, thanks to the rework command, which is similar to add:

> git checkout -b mastodon
> sqitch rework userflips -n 'Adds userflips.mastodon.'
Added "userflips [userflips@v1.0.0-dev2]" to sqitch.plan.
Modify these files as appropriate:
  * deploy/userflips.sql
  * revert/userflips.sql
  * verify/userflips.sql

Oh, so we can edit those files in place. Nice! How does Sqitch do it? Well, in point of fact, it has copied the files to stand in for the previous instance of the userflips change, which we can see via git status:

> git status
On branch mastodon
Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
        modified:   revert/userflips.sql
        modified:   sqitch.plan

Untracked files:
  (use "git add <file>..." to include in what will be committed)
        deploy/userflips@v1.0.0-dev2.sql
        flipr-1.0.0-dev2/
        revert/userflips@v1.0.0-dev2.sql
        verify/userflips@v1.0.0-dev2.sql

no changes added to commit (use "git add" and/or "git commit -a")

The “untracked files” part of the output is the first thing to notice. They are all named userflips@v1.0.0-dev2.sql. What that means is: “the userflips change as it was implemented as of the @v1.0.0-dev2 tag.” These are copies of the original scripts, and thereafter Sqitch will find them when it needs to run scripts for the first instance of the userflips change. As such, it’s important not to change them again. But hey, if you’re reworking the change, you shouldn’t need to.

The other thing to notice is that revert/userflips.sql has changed. Sqitch replaced it with the original deploy script. As of now, deploy/userflips.sql and revert/userflips.sql are identical. This is on the assumption that the deploy script will be changed (we’re reworking it, remember?), and that the revert script should actually change things back to how they were before.

Fortunately, our view deploy scripts are already almost idempotent – that is, able to be applied multiple times without changing the result beyond the initial application. If it’s not, you will likely need to modify it so that it properly restores things to how they were after the original deploy script was deployed. Or, more simply, it should revert changes back to how they were as-of the deployment of deploy/userflips@v1.0.0-dev2.sql.

Fortunately, our view deploy scripts are already idempotent, thanks to the use of the OR REPLACE expression. No matter how many times a deployment script is run, the end result will be the same instance of the view, with no duplicates or errors.

As a result, there is no need to explicitly add changes. So go ahead. Modify deploy/userflips.sql to add the mastodon column.

--- a/deploy/userflips.sql
+++ b/deploy/userflips.sql
@@ -3,6 +3,6 @@
 -- requires: flips

 CREATE OR REPLACE VIEW userflips AS
-SELECT f.id, u.nickname, u.fullname, f.body, f.created_at AS created_at
+SELECT f.id, u.nickname, u.fullname, u.mastodon, f.body, f.created_at created_at
   FROM users u
   JOIN flips f ON u.nickname = f.nickname;

Next, modify verify/userflips.sql to check for the mastodon column. Here’s the diff:

--- a/verify/userflips.sql
+++ b/verify/userflips.sql
@@ -1,5 +1,5 @@
 -- Verify flipr:userflips on clickhouse

-SELECT id, nickname, fullname, body, created_at
+SELECT id, nickname, mastodon, fullname, body, created_at
   FROM userflips
  WHERE false;

Now try a deployment:

> sqitch deploy
Deploying changes to flipr_test
  + userflips .. ok

So, are the changes deployed?

> clickhouse client -d flipr_test -q 'describe userflips'
id      Int32
nickname        String
fullname        String
mastodon        String
body    String
created_at      DateTime64(6, \'UTC\')

Awesome, the view now includes the mastodon column. But can we revert?

> sqitch revert --to @HEAD^ -y
Reverting changes to hashtags @v1.0.0-dev2 from flipr_test
  - userflips .. ok

Did that work, is the mastodon column gone?

> clickhouse client -d flipr_test -q 'describe userflips'
id      Int32
nickname        String
fullname        String
body    String
created_at      DateTime64(6, \'UTC\')

Yes, it works! Sqitch properly finds the original instances of these changes in the new script files that include tags.

Excellent. Let’s go ahead and commit these changes and merge them into main:

> rm -rf flipr-1.0.0-dev2
> git add .
> git commit -m 'Add the mastodon column to the userflips view.'
[mastodon 571662b] Add the mastodon column to the userflips view.
 7 files changed, 26 insertions(+), 4 deletions(-)
 create mode 100644 deploy/userflips@v1.0.0-dev2.sql
 create mode 100644 revert/userflips@v1.0.0-dev2.sql
 create mode 100644 verify/userflips@v1.0.0-dev2.sql
> git checkout main
> git merge --no-ff mastodon -m "Merge branch 'mastodon'"
Merge made by the 'ort' strategy.
 deploy/userflips.sql             | 2 +-
 deploy/userflips@v1.0.0-dev2.sql | 8 ++++++++
 revert/userflips.sql             | 9 +++++++--
 revert/userflips@v1.0.0-dev2.sql | 3 +++
 sqitch.plan                      | 1 +
 verify/userflips.sql             | 2 +-
 verify/userflips@v1.0.0-dev2.sql | 5 +++++
 7 files changed, 26 insertions(+), 4 deletions(-)
 create mode 100644 deploy/userflips@v1.0.0-dev2.sql
 create mode 100644 revert/userflips@v1.0.0-dev2.sql
 create mode 100644 verify/userflips@v1.0.0-dev2.sql

More to Come

Sqitch is a work in progress. Better integration with version control systems is planned to make managing idempotent reworkings even easier. Stay tuned.

Author

• David E. Wheeler david@justatheory.com

License

Copyright (c) 2012-2025 David E. Wheeler, 2012-2021 iovation Inc.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.