Rollbacks Are Normal

  • Comments posted to this topic are about the item Rollbacks Are Normal

  • I really like this article and the Google blog post you mentioned. It makes sense, a lot of sense. It's brilliant, in fact.

    But I think you made a mistake. The Google blog post didn't suggestion rolling back the change in the database schema. Quoting from the blog post:

    Now, if there are any problems with either of the new binaries then you can roll back to a previous version without having to also roll back the schema.

    Using the Google blog posts terminology, version v + 1 of an app takes into account only the changes to the database schema, without adding any new functionality to version v. Then you apply the database schema changes, and then introduce the functionality changes to the app in version v + 2.

    However, you bring up something I don't think was necessarily taken into account in Google's blog post. That's how might things be effected if you drop some database object, like a table, view or stored proc? It almost seems to me as though Google's post covered only adding something to a database schema, rather than removing something big like dropping an object.

    Still, I really like Google's approach on this and it certainly covers what likely is the majority of database schema changes.

    Kindest Regards, Rod Connect with me on LinkedIn.

  • I assumed that v+1 should maintain the functionality without that object, which would be consistent with the new schema. I'm not sure how you do that, maybe that means you never have an upgrade that adds a new object and drops an old one that are mutually exclusive, but that you need to maintain the old object through at least one upgrade version before you can comfortably drop it?

    -------------------------------------------------------------------------------------------------------------------------------------
    Please follow Best Practices For Posting On Forums to receive quicker and higher quality responses

  • I don't know either, Jonathan. to me, dropping an object has bigger considerations to take into account. Perhaps it doesn't if the data is refactored into other objects. But if you're just doing away with something, I don't see how it can't have an impact even upon the v+1 version of the application.

    Kindest Regards, Rod Connect with me on LinkedIn.

  • I did this recently when I needed to fix a naming convention, maybe something like this would work for v+1(very simplistic):

    1. copy records in oldTable into newTable (with correct naming convention), both objects still existing (v+1).
    2. create view with same name as oldTable
    3. Drop oldTable (except we'd skip this step in the software version v+1)
    4. refactor code to point to newTable, where it references oldTable (which is really now a view)
    5. Once done, drop oldTable view. (this would happen in v+2)

    -------------------------------------------------------------------------------------------------------------------------------------
    Please follow Best Practices For Posting On Forums to receive quicker and higher quality responses

  • Regardless of what formal process you follow and which tools you use for the delivery mechanism, many of us have known for years that it's best to perform iterative deployments.

    "Do not seek to follow in the footsteps of the wise. Instead, seek what they sought." - Matsuo Basho

  • Eric M Russell - Thursday, September 20, 2018 11:44 AM

    Regardless of what formal process you follow and which tools you use for the delivery mechanism, many of us have known for years that it's best to perform iterative deployments.

    For me, the example given of how Google does it, was very helpful. I've heard "iterative deployments", but I have never actually seen it in action.

    Kindest Regards, Rod Connect with me on LinkedIn.

  • jonathan.crawford - Thursday, September 20, 2018 10:56 AM

    I did this recently when I needed to fix a naming convention, maybe something like this would work for v+1(very simplistic):

    1. copy records in oldTable into newTable (with correct naming convention), both objects still existing (v+1).
    2. create view with same name as oldTable
    3. Drop oldTable (except we'd skip this step in the software version v+1)
    4. refactor code to point to newTable, where it references oldTable (which is really now a view)
    5. Once done, drop oldTable view. (this would happen in v+2)

    I would just use synonyms.

    Drew

    J. Drew Allen
    Business Intelligence Analyst
    Philadelphia, PA

  • Rod at work - Thursday, September 20, 2018 1:28 PM

    Eric M Russell - Thursday, September 20, 2018 11:44 AM

    Regardless of what formal process you follow and which tools you use for the delivery mechanism, many of us have known for years that it's best to perform iterative deployments.

    For me, the example given of how Google does it, was very helpful. I've heard "iterative deployments", but I have never actually seen it in action.

    By "iterative deployment", I simply mean that a major release to an existing production system is deployed incrementally in units, and there could be extended periods of time and testing between each deployment. Components that are functionally dependent (ie: adding a column to table and updating the views and forms that use the column) are deployed within the same unit. From my experience across multiple employers, it's very common.

    "Do not seek to follow in the footsteps of the wise. Instead, seek what they sought." - Matsuo Basho

  • Rod at work - Wednesday, September 19, 2018 9:16 AM

    I really like this article and the Google blog post you mentioned. It makes sense, a lot of sense. It's brilliant, in fact.

    But I think you made a mistake. The Google blog post didn't suggestion rolling back the change in the database schema. Quoting from the blog post:

    Now, if there are any problems with either of the new binaries then you can roll back to a previous version without having to also roll back the schema.

    Using the Google blog posts terminology, version v + 1 of an app takes into account only the changes to the database schema, without adding any new functionality to version v. Then you apply the database schema changes, and then introduce the functionality changes to the app in version v + 2.

    However, you bring up something I don't think was necessarily taken into account in Google's blog post. That's how might things be effected if you drop some database object, like a table, view or stored proc? It almost seems to me as though Google's post covered only adding something to a database schema, rather than removing something big like dropping an object.

    Still, I really like Google's approach on this and it certainly covers what likely is the majority of database schema changes.

    True, but if the issue is with the database, you can roll back the schema changes without breaking the app. You're also not under pressure to roll back changes in the db because users must wait. You can do this in a more controlled fashion.

  • jonathan.crawford - Wednesday, September 19, 2018 10:06 AM

    I assumed that v+1 should maintain the functionality without that object, which would be consistent with the new schema. I'm not sure how you do that, maybe that means you never have an upgrade that adds a new object and drops an old one that are mutually exclusive, but that you need to maintain the old object through at least one upgrade version before you can comfortably drop it?

    Use time to your advantage. Never add and drop the objects in the same release. Add, stabilize, drop in a future release.

Viewing 11 posts - 1 through 10 (of 10 total)

You must be logged in to reply to this topic. Login to reply