Allow SQL interface to be enabled without "experimental" compat flag. #2666
Conversation
kentonv
force-pushed
the
kenton/enable-sql
branch
from
66070e3 to
de7e109
Compare
Instead, it can be enabled in the workerd config. In production, it'll be enabled implicitly if the new SQLite-based storage engine is in use. But `workerd` running locally always uses SQLite, even when testing code intended to run against non-SQLite namespaces in production, so we need a config option for workerd.
Point-in-time recovery is only supported for SQLite-backed DOs in production; PITR methods will throw for non-SQLite namespaces. abort() will work for all DOs. I think we've generally agreed this should be made public, but it's especially needed with the PITR APIs since you need a way to restart the object immediately after calling `onNextSessionRestoreBookmark()`.
kentonv
force-pushed
the
kenton/enable-sql
branch
from
de7e109 to
fbf8de8
Compare
| @@ -579,6 +579,14 @@ struct Worker { | |||
| # pinned to memory forever, so we provide this flag to change the default behavior. | |||
| # | |||
| # Note that this is only supported in Workerd; production Durable Objects cannot toggle eviction. | |||
|
|
|||
| enableSql @4 :Bool; | |||
There was a problem hiding this comment.
I understand how this field would be in a union with ephemeralLocal but it does seem like enableSql would be compatible with preventEviction and uniqueKey. Are we modeling the config this way because those combinations haven't been tested?
There was a problem hiding this comment.
Also, it appears we are manually copying this file over here(workers-sdk)? 😕
There was a problem hiding this comment.
This field isn't a member of a union. The union ended on line 574.
Also, it appears we are manually copying this file over here(workers-sdk)? 😕
Manual copying is a pretty normal way to share schemas. It's not a big deal, you just overwrite the copy with the updated version if and when you want to change workers-sdk to use the new field.
|
After some discussion I think I'm going to update this PR so that instead of making |
|
Updated as described: Now |
Having `storage.sql` be optional is potentially annoying for two reasons: - TypeScript will force people to check if it's present, even though apps that have configured it should be able to expect it's always present. - We'd like to provide a more detailed error message telling people how to configure SQL. So, this commit changes things so `storage.sql` is always present, but its methods will throw exceptions if the DO isn't SQLite-backed.
kentonv
force-pushed
the
kenton/enable-sql
branch
from
6201cfb to
4ae0e11
Compare
Instead, it can be enabled in the workerd config.
In production, it'll be enabled implicitly if the new SQLite-based storage engine is in use. But
workerdrunning locally always uses SQLite, even when testing code intended to run against non-SQLite namespaces in production, so we need a config option for workerd.Also enable PITR and abort() APIs without "experimental".
Point-in-time recovery is only supported for SQLite-backed DOs in production; PITR methods will throw for non-SQLite namespaces.
abort() will work for all DOs. I think we've generally agreed this should be made public, but it's especially needed with the PITR APIs since you need a way to restart the object immediately after calling
onNextSessionRestoreBookmark().