The documentation is clear and the answer seems to be obvious. However, as I'm discussing this matter quite frequently, it turns out that some people need a small clarification. I have decided that today is the last day when I explain this issue to anyone. From now on, I will be sending a link to this article and feel exempted from further explanations.
Rescue which does not work as you wish
The following example does not roll back your post object.
ActiveRecord::Base.transaction do
post.update!(title: "cow")
post.update!(content: "beer")
raise StandardError, "cow and beer are not compatible"
rescue StandardError => e
Rails.logger.error(e.message)
end
The error is rescued and handled. It is not propagated beyond the rescue block. As a result, it doesn't affect the transaction block.
Example
As a quick example, let's create a post with a title and some content. In a transaction block, it is updated twice and then an error is raised. The error is rescued and you can clearly see a COMMIT statement in the SQL log. You can also read the updated values from the post object.
If you intended it to roll back the transaction - it just does not work like that.
> post = Post.create(title: 'bee', content: 'honey')
TRANSACTION (0.1ms) BEGIN
Post Create (0.6ms) INSERT INTO "posts"("title", "content", "created_at", "updated_at") VALUES ($1, $2, $3, $4) RETURNING "id" [["title", "bee"], ["content", "honey"], ["created_at", "2022-09-19 13:52:12.518986"], ["updated_at", "2022-09-19 13:52:12.518986"]]
TRANSACTION (2.0ms) COMMIT
=> #<Post:0x0000000112f4ea30 id: 3, title: "bee", content: "honey", created_at: Mon, 19 Sep 2022 13:52:12.518986000 UTC +00:00, updated_at: Mon, 19 Sep 2022 13:52:12.518986000 UTC +00:00>
> ActiveRecord::Base.transaction do
> post.update!(title: "cow")
> post.update!(content: "beer")
>
> raise StandardError, "cow and beer are not compatible"
> rescue StandardError => e
> Rails.logger.error(e.message)
> end
TRANSACTION (0.1ms) BEGIN
Post Update (0.3ms) UPDATE "posts" SET "title" = $1, "updated_at" = $2 WHERE "posts"."id" = $3 [["title", "cow"], ["updated_at", "2022-09-19 13:52:17.956795"], ["id", 3]]
Post Update (0.2ms) UPDATE "posts" SET "content" = $1, "updated_at" = $2 WHERE "posts"."id" = $3 [["content", "beer"], ["updated_at", "2022-09-19 13:52:17.958060"], ["id", 3]]
cow and beer are not compatible
TRANSACTION (2.0ms) COMMIT
=> 32
> post.reload
Post Load (0.3ms) SELECT "posts".* FROM "posts" WHERE "posts"."id" = $1 LIMIT $2 [["id", 3], ["LIMIT", 1]]
=> #<Post:0x0000000112f4ea30 id: 3, title: "cow", content: "beer", created_at: Mon, 19 Sep 2022 13:52:12.518986000 UTC +00:00, updated_at: Mon, 19 Sep 2022 13:52:17.958060000 UTC +00:00>
A working example of a transaction rollback
begin
ActiveRecord::Base.transaction do
post.update!(title: "cow")
post.update!(content: "beer")
raise StandardError, "cow and beer are not compatible"
end
rescue StandardError => e
Rails.logger.error(e.message)
end
In this case, the error affects the transaction block and stops it from being committed. Later, the error is cached by the rescue instruction and handled appropriately.
Example
Let's use a similar example. In the transaction block, we update the post twice and then raise an error. The error is rescued outside of the transaction and you can see the ROLLBACK statement in the SQL log. You can also read the final values from the post object and see they were not changed.
If you intended it to roll back the transaction - congratulations, well done!
> post = Post.create(title: 'bee', content: 'honey')
TRANSACTION (0.1ms) BEGIN
Post Create (0.2ms) INSERT INTO "posts" ("title", "content", "created_at", "updated_at") VALUES ($1, $2, $3, $4) RETURNING "id" [["title", "bee"], ["content", "honey"], ["created_at", "2022-09-19 13:59:45.933973"], ["updated_at", "2022-09-19 13:59:45.933973"]]
TRANSACTION (2.1ms) COMMIT
=> #<Post:0x0000000112d4f298 id: 4, title: "bee", content: "honey", created_at: Mon, 19 Sep 2022 13:59:45.933973000 UTC +00:00, updated_at: Mon, 19 Sep 2022 13:59:45.933973000 UTC +00:00>
> begin
> ActiveRecord::Base.transaction do
> post.update!(title: "cow")
> post.update!(content: "beer")
>
> raise StandardError, "cow and beer are not compatible"
> end
> rescue StandardError => e
> Rails.logger.error(e.message)
> end
TRANSACTION (0.2ms) BEGIN
Post Update (0.4ms) UPDATE "posts" SET "title" = $1, "updated_at" = $2 WHERE "posts"."id" = $3 [["title", "cow"], ["updated_at", "2022-09-19 13:59:53.937241"], ["id", 4]]
Post Update (0.2ms) UPDATE "posts" SET "content" = $1, "updated_at" = $2 WHERE "posts"."id" = $3 [["content", "beer"], ["updated_at", "2022-09-19 13:59:53.939197"], ["id", 4]]
TRANSACTION (0.1ms) ROLLBACK
cow and beer are not compatible
=> 32
> post.reload
Post Load (0.3ms) SELECT "posts".* FROM "posts" WHERE "posts"."id" = $1 LIMIT $2 [["id", 4], ["LIMIT", 1]]
=> #<Post:0x0000000112d4f298 id: 4, title: "bee", content: "honey", created_at: Mon, 19 Sep 2022 13:59:45.933973000 UTC +00:00, updated_at: Mon, 19 Sep 2022 13:59:45.933973000 UTC +00:00>
Is rescuing within the block ever correct?
Definitely. There are numerous use cases where it is the correct approach. It allows you to ignore some of the errors and let the transaction commit. You may have additional logic in such a rescue block to decide the fate of the transaction.
But please, use it wisely. If you have any doubts, check the logs for TRANSACTION COMMIT or ROLLBACK entries.
Happy Coding!
Check my Twitter
Check my Linkedin
