"Course completion" was causing browser crashes when it attempted to display
10k course items (which is the maximum allowed). Displaying 10000 bars does not
make sense anyway, so we threshold this number to 100.
Superset was ignoring the cairn-redis container, so we decided to remove it.
But the superset celery worker was writing to the /0 database, which conflicts
with the default celery database from the lms/cms. We fix this issue by running
in a different database.
The CELERY_RESULT_BACKEND seemed to be completely useless, so we got rid of it.
Vector is spamming us with messages concerning absent containers. Such messages include:
tutor_local-cairn-vector-1 | 2022-10-07T13:09:27.053010Z INFO source{component_kind="source" component_id=docker_logs component_type=docker_logs component_name=docker_logs}: vector::internal_events::docker_logs: Started watching for container logs. container_id=680c2bf183676caadee2c165b26493f90fb5d02cb15b9f2579a106fdd436f827
Or:
{“log”:“2022-10-07T07:19:17.839042Z INFO vector::internal_events::docker_logs: Started watching for container logs. container_id=a932d5916f9a1d1848224b71070591d16c813e83796d6850fa64b0a5f956d5ed\n”,“stream”:“stderr”,“time”:“2022-10-07T07:19:17.83961548Z”}
{“log”:“2022-10-07T07:19:17.839311Z INFO vector::internal_events::docker_logs: Stopped watching for container logs. container_id=cb9a4330dc154e02f02f6f6b49d9ddb04b5357083becdf6e9c5cbf49f4d01f68\n”,“stream”:“stderr”,“time”:“2022-10-07T07:19:17.83967086Z”}
See also: https://github.com/vectordotdev/vector/issues/8881
This issue has affected multiple users in the past. To address it, we first
tried to upgrade Vector, which did not help. Thus, we decided to reduce the
verbosity of Vector.
When creating a new user, but with an email address already associated
to a different user, the `cairn createuser` script crashes with an
obscure error. This slightly improves the error messsage.
See: https://discuss.openedx.org/t/cairn-password-reset/8226
This is necessary for some customers who are not using the
common.djangoapps.track.shim.LegacyFieldMappingProcessor in
settings.EVENT_TRACKING_BACKENDS.
We could not create datasets directly from physical tables. Turns out, this is
caused by the fact that reflection is enabled on the clickhouse databases in
Superset. To disable reflection, we manually specify the server version.
In the future, we must remove the server version from the database, and instead
specify engine_reflection=no in the DSN. We will be able to do that once a new
version of clickhouse-sqlalchemy has been released.
See:
https://github.com/xzkostyan/clickhouse-sqlalchemy/issues/151https://github.com/xzkostyan/clickhouse-sqlalchemy/issues/140
This is preventing us from creating datasets from physical tables.
