Avoid version collisions #58
Conversation
|
closes #45 |
|
Cassandra is starting too slowly, and is not available when the tests are running, I'll figure out something. Ideas are welcome. |
|
done |
|
Hi @gravis I was wondering why this used auto-incremented numbers when Rails' Activerecord migrations had moved to timestamps long ago. 👍 for this PR, hopefully it gets merged soon. |
|
Can we fix the tests? Happy to merge this. |
|
I have no idea why Cassandra is failing with: I'll try to add more time to wait for startup |
|
👍 |
This change insert a timestamped version into schema migration table, instead of incrementing a field. That way, we have a history of passed migrations. Next step: Apply all pending migrations. Pending migrations being the list of all migrations from files, except the ones already applied (d.Versions())
sleeping is unrelevent here, the docker-compose command is launched after the sleep
gravis
force-pushed
the
avoid-version-collisions
branch
from
7934ec3
to
b981eae
Compare
|
@mattes Cheers, |
|
We really need this merged, please take a look @mattes |
|
@raphaeljlps you can use our fork until that: https://github.com/gemnasium/migrate |
|
@gravis thank you, thats just what I was going to do |
|
@gravis is there a way to migrate old migration files ? |
|
Unfortunately, no :( |
|
Prompted by comments on #95, I've ported to gemnasium/migrate fork the ability to be backwards compatible with mattes/migrate. See gemnasium/migrate#3 |
|
waiting for this to be merged, though already started using @gravis fork... |
|
Any updates on this? |
Don't use incremental integers, as it will fail with large teams. This PR is using timestamps instead of versions, like ruby on rails does since a few years now, the same reasons.
"Earlier" migrations will be processed when a branch is merged. ie: if 20060102150405 and 20060102150406 have been applied already, 20060102150403 can be applied afterwards.
The migration will fail if 2 versions have the same id, this was already the case.
Note that the schema changed for some DB, moving from
inttobigint. I also have removed the counter column from cassandra (and a lot of code based on booleans).