Name¶
sqitchtutorial-sqlite - A tutorial introduction to Sqitch change management on
SQLite
Synopsis¶
sqitch *
Description¶
This tutorial explains how to create a sqitch-enabled SQLite 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 new project from scratch, a fictional antisocial
networking site called Flipr. All examples use Git <
http://git-scm.com/>
as the VCS and SQLite <
http://www.sqlite.org/> as the storage engine.
If you'd like to manage an PostgreSQL database, see sqitchtutorial.
If you'd like to manage an Oracle database, see sqitchtutorial-oracle.
If you'd like to manage an MySQL database, see sqitchtutorial-mysql.
If you'd like to manage an Firebird database, see sqitchtutorial-firebird.
If you'd like to manage an Vertica database, see sqitchtutorial-vertica.
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 -m 'Initialize project, add README.'
[master (root-commit) 253542e] Initialize project, add README.
1 file changed, 37 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
<
https://github.com/theory/sqitch-sqlite-intro>.
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, so let's specify one when we initialize Sqitch:
> sqitch --engine sqlite init flipr --uri https://github.com/theory/sqitch-sqlite-intro/
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 = sqlite
# plan_file = sqitch.plan
# top_dir = .
# deploy_dir = deploy
# revert_dir = revert
# verify_dir = verify
# extension = sql
# [core "sqlite"]
# target = db:sqlite:
# registry = sqitch
# client = sqlite3
Good, it picked up on the fact that we're creating changes for the SQLite
engine, thanks to the "--engine sqlite" option, and saved it to the
file. Furthermore, it wrote a commented-out "[core
"sqlite"]" section with all the available SQLite
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 SQLite's "sqlite3" client is not in the path on my
system, let's go ahead an tell it where to find the client on our computer.
> sqitch config --user core.sqlite.client /opt/local/bin/sqlite3
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. OXVera'
> sqitch config --user user.email 'marge@example.com'
Have a look at
~/.sqitch/sqitch.conf and you'll see this:
> cat ~/.sqitch/sqitch.conf
[core "sqlite"]
client = /opt/local/bin/sqlite3
[user]
name = Marge N. OXVera
email = marge@example.com
Which means that Sqitch should be able to find "sqlite3" 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-b2
%project=flipr
%uri=https://github.com/theory/sqitch-sqlite-intro/
Note that it has picked up on 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 -m 'Initialize Sqitch configuration.'
[master 91e2f0d] Initialize Sqitch configuration.
2 files changed, 19 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 users
BEGIN;
-- XXX Add DDLs here.
COMMIT;
What we want to do is to replace the "XXX" comment with the
"CREATE TABLE" statement, like so:
-- Deploy users
BEGIN;
CREATE TABLE users (
nickname TEXT PRIMARY KEY,
password TEXT NOT NULL,
fullname TEXT NOT NULL,
twitter TEXT NOT NULL,
timestamp DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP
);
COMMIT;
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 users
BEGIN;
DROP TABLE users;
COMMIT;
Now we can try deploying this change:
> sqitch deploy db:sqlite:flipr_test.db
Adding registry tables to db:sqlite:sqitch.db
Deploying changes to db:sqlite:flipr_test.db
+ users .. ok
First Sqitch created the registry database and tables used to track database
changes. The registry is separate from the database to which the
"users" change was deployed; by default, its name is
"sqitch.$suffix", where $suffix is the same as the suffix on the
target database, if any. It lives in the same directory as the target
database. This will be useful if you use the SQLite "ATTACHDATABASE"
<
http://www.sqlite.org/lang_attach.html> command to manage multiple
database files in a single project. In that case, you will want to use the
same file for all the databases. Keep them all in the same directory with the
same suffix and you get just that with the default sqitch database.
If you'd like it to have a different name for the registry database, use
"sqitch config core.sqlite.registry $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, or if you
do, but they live in different directories.
Next, Sqitch deploys changes to the target database, which we specified on the
command-line. We only have one 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 schema:
> sqlite3 flipr_test.db '.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, twitter
FROM users
WHERE 0;
Now you can run the "verify" script with the "verify"
command:
> sqitch verify db:sqlite:flipr_test.db
Verifying db:sqlite:flipr_test.db
* 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, timestamp
FROM users_nonesuch
WHERE 0;
Then "verify" again:
> sqitch verify db:sqlite:flipr_test.db
Verifying db:sqlite:flipr_test.db
* users .. Error: near line 5: no such table: users_nonesuch
# Verify script "verify/users.sql" failed.
not ok
Verify Summary Report
---------------------
Changes: 1
Errors: 1
Verify failed
SQLite 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 tables from the
registry database:
> sqitch status db:sqlite:flipr_test.db
# On database db:sqlite:flipr_test.db
# Project: flipr
# Change: f30fe47f5f99501fb8d481e910d9112c5ac0a676
# Name: users
# Deployed: 2013-12-31 10:26:59 -0800
# By: Marge N. OXVera <marge@example.com>
#
Nothing to deploy (up-to-date)
Let's make sure that we can revert the change:
> sqitch revert db:sqlite:flipr_test.db
Revert all changes from db:sqlite:flipr_test.db? [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:
> sqlite3 flipr_test.db '.tables'
And the status message should reflect as much:
> sqitch status db:sqlite:flipr_test.db
# On database db:sqlite:flipr_test.db
No changes deployed
Of course, since nothing is deployed, the "verify" command has nothing
to verify:
> sqitch verify db:sqlite:flipr_test.db
Verifying db:sqlite:flipr_test.db
No changes deployed
However, we still have a record that the change happened, visible via the
"log" command:
> sqitch log db:sqlite:flipr_test.db
On database db:sqlite:flipr_test.db
Revert f30fe47f5f99501fb8d481e910d9112c5ac0a676
Name: users
Committer: Marge N. OXVera <marge@example.com>
Date: 2013-12-31 10:53:25 -0800
Creates table to track our users.
Deploy f30fe47f5f99501fb8d481e910d9112c5ac0a676
Name: users
Committer: Marge N. OXVera <marge@example.com>
Date: 2013-12-31 10:26:59 -0800
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. Let's tell Git to ignore
*.db files and then commit it.
> echo '*.db' > .gitignore
> git add .
> git commit -m 'Add users table.'
[master 6725454] Add users table.
5 files changed, 31 insertions(+)
create mode 100644 .gitignore
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 db:sqlite:flipr_test.db --verify
Deploying changes to db:sqlite:flipr_test.db
+ users .. ok
And now the "users" table should be back:
> sqlite3 flipr_test.db '.tables'
users
When we look at the status, the deployment will be there:
> sqitch status db:sqlite:flipr_test.db
# On database db:sqlite:flipr_test.db
# Project: flipr
# Change: f30fe47f5f99501fb8d481e910d9112c5ac0a676
# Name: users
# Deployed: 2013-12-31 10:57:55 -0800
# By: Marge N. OXVera <marge@example.com>
#
Nothing to deploy (up-to-date)
On Target¶
I'm getting a little tired of always having to type
"db:sqlite:flipr_test.db", aren't you? This database connection URI
<
https://github.com/theory/uri-db/> 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:sqlite:flipr_test.db
The "target" command, inspired by "git-remote"
<
http://git-scm.com/docs/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 config core.sqlite.target 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: f30fe47f5f99501fb8d481e910d9112c5ac0a676
# Name: users
# Deployed: 2013-12-31 10:57:55 -0800
# By: Marge N. OXVera <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 target and always verify.'
[master 5fb57ec] Set default 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 flips
-- requires: users
BEGIN;
CREATE TABLE flips (
id INTEGER PRIMARY KEY AUTOINCREMENT,
nickname TEXT NOT NULL REFERENCES users(nickname),
body TEXT NOT NULL DEFAULT '' CHECK ( length(body) <= 180 ),
timestamp DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP
);
COMMIT;
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 references the "users" table.
This is why we need to require the "users" change.
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 flips
BEGIN;
SELECT id, nickname, body, timestamp
FROM flips
WHERE 0;
COMMIT;
Now for the revert script: all we have to do is drop the table. Add this to
revert/flips.sql:
-- Revert flips
BEGIN;
DROP TABLE flips;
COMMIT;
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:
> sqlite3 flipr_test.db '.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: 32ee57069c0d7fec52b6f86f453dc0c16bc1090a
# Name: flips
# Deployed: 2013-12-31 11:02:51 -0800
# By: Marge N. OXVera <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:
> sqlite3 flipr_test.db '.tables'
users
The "status" command politely informs us that we have undeployed
changes:
> sqitch status
# On database flipr_test
# Project: flipr
# Change: f30fe47f5f99501fb8d481e910d9112c5ac0a676
# Name: users
# Deployed: 2013-12-31 10:57:55 -0800
# By: Marge N. OXVera <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.'
[master 21cba95] Add flips table.
4 files changed, 30 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: 32ee57069c0d7fec52b6f86f453dc0c16bc1090a
# Name: flips
# Deployed: 2013-12-31 11:05:44 -0800
# By: Marge N. OXVera <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 VIEW userflips AS
SELECT f.id, u.nickname, u.fullname, f.body, f.timestamp
FROM users u
JOIN flips f ON u.nickname = f.nickname;
Add this SQL to
verify/userflips.sql
SELECT id, nickname, fullname, body, timestamp
FROM userflips
WHERE 0;
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.'
[master c74bfb4] Add the userflips view.
4 files changed, 29 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.'
[master 7a479fd] Tag the database with v1.0.0-dev1.
1 file changed, 1 insertion(+)
> git tag v1.0.0-dev1 -am 'Tag v1.0.0-dev1'
We can try deploying to make sure the tag gets picked up like so:
> mkdir dev
> sqitch deploy db:sqlite:dev/flipr.db
Adding registry tables to db:sqlite:dev/sqitch.db
Deploying changes to db:sqlite:dev/flipr.db
+ users ................... ok
+ flips ................... ok
Great, both changes were deployed and "userflips" was tagged with
"@v1.0.0-dev1". Let's have a look at the status:
> sqitch status db:sqlite:dev/flipr_dev.db
# On database db:sqlite:dev/flipr_dev.db
# Project: flipr
# Change: 60ee3aba0445bf3287f9dc1dd97b1877523fa139
# Name: userflips
# Tag: @v1.0.0-dev1
# Deployed: 2013-12-31 11:19:15 -0800
# By: Marge N. OXVera <marge@example.com>
#
Nothing to deploy (up-to-date)
Note the listing of the tag as part of the status message. Now let's bundle
everything up for release:
> rm -rf dev
> 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. Let's try
deploying it:
> cd bundle
> sqitch deploy db:sqlite:flipr_prod.db
Adding registry tables to db:sqlite:sqitch.db
Deploying changes to db:sqlite:flipr_prod.db
+ users ................... ok
+ flips ................... ok
+ userflips @v1.0.0-dev1 .. ok
Looks much the same as before, eh? Package it up and ship it!
> rm *.db
> 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. But
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 REFERENCES flips(id),
hashtag TEXT NOT NULL CHECK ( length(hashtag) > 0 ),
PRIMARY KEY (flip_id, hashtag)
);
Again, select from the table in
verify/hashtags.sql:
SELECT flip_id, hashtag FROM hashtags WHERE 0;
And drop it in
revert/hashtags.sql
DROP TABLE hashtags;
And give it a whirl:
> sqitch deploy
Deploying changes to flipr_test
+ hashtags .. ok
Look good?
> sqitch status --show-tags
# On database flipr_test
# Project: flipr
# Change: 1352464e8b5f3d5eeac76a1986379f07de43bffd
# Name: hashtags
# Deployed: 2013-12-31 11:30:53 -0800
# By: Marge N. OXVera <marge@example.com>
#
# Tag:
# @v1.0.0-dev1 - 2013-12-31 11:13:49 -0800 - Marge N. OXVera <marge@example.com>
#
Nothing to deploy (up-to-date)
Note the use of "--show tags" to show all the deployed tags. Make sure
we can revert, too:
> sqitch revert --to @HEAD^ -y
Reverting changes to userflips @v1.0.0-dev1 from flipr_test
- hashtags .. ok
> sqitch deploy
Deploying changes to flipr_test
+ hashtags .. ok
Great! Now make it so:
> git add .
> git commit -m 'Add hashtags table.'
[hashtags 94f02b8] Add hashtags table.
4 files changed, 28 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 "master".
Emergency¶
Let's do it:
> git checkout master
Switched to branch 'master'
> git pull
Updating 7a479fd..47a4107
Fast-forward
deploy/lists.sql | 13 +++++++++++++
revert/lists.sql | 7 +++++++
sqitch.plan | 2 ++
verify/lists.sql | 9 +++++++++
4 files changed, 31 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
"master". 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 master from the origin
should have raised a red flag. It's considered a bad practice not to look at
what's changed in "master" before merging in a branch. What one
should do is either:
- •
- Rebase the hashtags branch from master before merging. This
"rewinds" the branch changes, pulls from "master", and
then replays the changes back on top of the pulled changes.
- •
- Create a patch and apply that to master. 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 master:
> git reset --hard HEAD
HEAD is now at 47a4107 Merge branch 'lists'
That throws out our botched merge. Now let's go back to our branch and rebase it
on "master":
> git checkout hashtags
Switched to branch 'hashtags'
> git rebase master
First, rewinding head to replay your work on top of it...
Applying: Add hashtags table.
Using index info to reconstruct a base tree...
M sqitch.plan
Falling back to patching base and 3-way merge...
Auto-merging sqitch.plan
CONFLICT (content): Merge conflict in sqitch.plan
Failed to merge in the changes.
Patch failed at 0001 Add hashtags table.
The copy of the patch that failed is found in:
.git/rebase-apply/patch
When you have resolved this problem, run "git rebase --continue".
If you prefer to skip this patch, run "git rebase --skip" instead.
To check out the original branch and stop rebasing, run "git rebase --abort".
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
<
http://git-scm.com/docs/gitattributes#_built-in_merge_drivers>:
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 master
First, rewinding head to replay your work on top of it...
Applying: Add hashtags table.
Using index info to reconstruct a base tree...
M sqitch.plan
Falling back to patching base and 3-way merge...
Auto-merging sqitch.plan
Ah, that looks a bit better. Let's have a look at the plan:
> cat sqitch.plan
%syntax-version=1.0.0-b2
%project=flipr
%uri=https://github.com/theory/sqitch-sqlite-intro/
users 2013-12-31T18:06:04Z Marge N. OXVera <marge@example.com> # Creates table to track our users.
flips [users] 2013-12-31T19:01:40Z Marge N. OXVera <marge@example.com> # Adds table for storing flips.
userflips [users flips] 2013-12-31T19:11:11Z Marge N. OXVera <marge@example.com> # Creates the userflips view.
@v1.0.0-dev1 2013-12-31T19:13:02Z Marge N. OXVera <marge@example.com> # Tag v1.0.0-dev1.
lists [users] 2013-12-31T19:28:05Z Marge N. OXVera <marge@example.com> # Adds table for storing lists.
hashtags [flips] 2013-12-31T19:30:13Z Marge N. OXVera <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 4f93ac4] 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 "master":
> git checkout master
Switched to branch 'master'
> git merge --no-ff hashtags -m "Merge branch 'hashtags'"
Merge made by the 'recursive' strategy.
.gitattributes | 1 +
deploy/hashtags.sql | 12 ++++++++++++
revert/hashtags.sql | 7 +++++++
sqitch.plan | 1 +
verify/hashtags.sql | 7 +++++++
5 files changed, 28 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-b2
%project=flipr
%uri=https://github.com/theory/sqitch-sqlite-intro/
users 2013-12-31T18:06:04Z Marge N. OXVera <marge@example.com> # Creates table to track our users.
flips [users] 2013-12-31T19:01:40Z Marge N. OXVera <marge@example.com> # Adds table for storing flips.
userflips [users flips] 2013-12-31T19:11:11Z Marge N. OXVera <marge@example.com> # Creates the userflips view.
@v1.0.0-dev1 2013-12-31T19:13:02Z Marge N. OXVera <marge@example.com> # Tag v1.0.0-dev1.
lists [users] 2013-12-31T19:28:05Z Marge N. OXVera <marge@example.com> # Adds table for storing lists.
hashtags [flips] 2013-12-31T19:30:13Z Marge N. OXVera <marge@example.com> # Adds table for storing hashtags.
Much much better, a nice clean master 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.'
[master 7abfd9b] 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 Twitter 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
Product demands. The upshot is that we need to update the
"userflips" view, which is used for the feature in question, to
include the Twitter user names.
Normally, modifying views in database changes is a PITA
<
http://www.urbandictionary.com/define.php?term=pita>. You have to make
changes like these:
- 1.
- Copy deploy/userflips.sql to
deploy/userflips_twitter.sql.
- 2.
- Edit deploy/userflips_twitter.sql to drop and re-create the view
with the "twitter" column to the view.
- 3.
- Copy deploy/userflips.sql to revert/userflips_twitter.sql.
Yes, copy the original change script to the new revert change.
- 4.
- Add a "DROP VIEW" statement to
revert/userflips_twitter.sql.
- 5.
- Copy verify/userflips.sql to
verify/userflips_twitter.sql.
- 6.
- Modify verify/userflips_twitter.sql to include a check for the
"twiter" column.
- 7.
- Test the changes to make sure you can deploy and revert the
"userflips_twitter" 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":
> sqitch rework userflips -n 'Adds userflips.twitter.'
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 master
# Your branch is ahead of 'origin/master' by 4 commits.
# (use "git push" to publish your local commits)
#
# Changes not staged for commit:
# (use "git add <file>..." to update what will be committed)
# (use "git checkout -- <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
# 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. Of course, the original deploy script
won't be idempotent <
http://en.wikipedia.org/wiki/Idempotence> -- that
is, able to be applied multiple times without changing the result beyond the
initial application. It could be if SQLite supported "CREATE OR REPLACE
VIEW", but since it doesn't, we will have to edit the script to drop the
view before creating it. Or, more simply, it needs to be updated to revert
changes back to how they were as-of the deployment of
deploy/userflips@v1.0.0-dev2.sql.
Modify
deploy/userflips.sql to add the "twitter" column; in
fact, let's also add a "DROP VIEW IF EXISTS" statement, in case we
need to rework this change again in the future:
@@ -4,8 +4,9 @@
BEGIN;
+DROP VIEW IF EXISTS userflips;
CREATE VIEW userflips AS
-SELECT f.id, u.nickname, u.fullname, f.body, f.timestamp
+SELECT f.id, u.nickname, u.fullname, u.twitter, f.body, f.timestamp
FROM users u
JOIN flips f ON u.nickname = f.nickname;
Next, modify
verify/userflips.sql to check for the "twitter"
column. Here's the diff:
@@ -2,7 +2,7 @@
BEGIN;
-SELECT id, nickname, fullname, body, timestamp
+SELECT id, nickname, fullname, twitter, body, timestamp
FROM userflips
WHERE 0;
And finally, modify
revert/userflips@v1.0.0-dev2.sql to drop the view
before creating it:
@@ -4,6 +4,7 @@
BEGIN;
+DROP VIEW IF EXISTS userflips;
CREATE VIEW userflips AS
SELECT f.id, u.nickname, u.fullname, f.body, f.timestamp
FROM users u
Note that if we had included that statement when we originally created the
"userflips" change, we wouldn't have to change this file at all.
Now try a deployment:
> sqitch deploy
Deploying changes to flipr_test
+ userflips .. ok
So, are the changes deployed?
> sqlite3 flipr_test.db '.schema userflips'
CREATE VIEW userflips AS
SELECT f.id, u.nickname, u.fullname, u.twitter, f.body, f.timestamp
FROM users u
JOIN flips f ON u.nickname = f.nickname;
Awesome, the view now includes the "twitter" 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 "twitter" column gone?
> sqlite3 flipr_test.db '.schema userflips'
CREATE VIEW userflips AS
SELECT f.id, u.nickname, u.fullname, f.body, f.timestamp
FROM users u
JOIN flips f ON u.nickname = f.nickname;
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:
> git add .
> git commit -m 'Add the twitter column to the userflips view.'
[master 3eb96d9] Add the twitter column to the userflips view.
7 files changed, 40 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) 2012-2014 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.