Commit Graph

2 Commits (2675a682184a1d7f7a29e4569a10e605fb549fd7)

Author SHA1 Message Date
Emel Şimşek 2675a68218
Make coordinator always in metadata by default in regression tests. (#6847)
DESCRIPTION: Changes the regression test setups adding the coordinator
to metadata by default.

When creating a Citus cluster, coordinator can be added in metadata
explicitly by running `citus_set_coordinator_host ` function. Adding the
coordinator to metadata allows to create citus managed local tables.
Other Citus functionality is expected to be unaffected.

This change adds the coordinator to metadata by default when creating
test clusters in regression tests.

There are 3 ways to run commands in a sql file (or a schedule which is a
sequence of sql files) with Citus regression tests. Below is how this PR
adds the coordinator to metadata for each.

1. `make <schedule_name>`
Changed the sql files (sql/multi_cluster_management.sql and
sql/minimal_cluster_management.sql) which sets up the test clusters such
that they call `citus_set_coordinator_host`. This ensures any following
tests will have the coordinator in metadata by default.
 
2. `citus_tests/run_test.py <sql_file_name>`
Changed the python code that sets up the cluster to always call `
citus_set_coordinator_host`.
For the upgrade tests, a version check is included to make sure
`citus_set_coordinator_host` function is available for a given version.

3. ` make check-arbitrary-configs  `     
Changed the python code that sets up the cluster to always call
`citus_set_coordinator_host `.

#6864 will be used to track the remaining work which is to change the
tests where coordinator is added/removed as a node.
2023-04-17 14:14:37 +03:00
Jelte Fennema 8866d9ac32
Reduce setup time of check-minimal and check-minimal-mx (#6117)
This change reduces the setup time of our minimal schedules in two ways:
1. Don't run `multi_cluster_managament`, but instead run a much smaller
   sql file with almost the same results. `multi_cluster_management`
   adds and removes lots of nodes and tests all kinds of failure
   scenarios. This is not needed for the minimal schedules. The only
   reason we were using it there was to get a working cluster of the
   layout that the tests expected. The new `minimal_cluster_management`
   test achieves this with much less work, going from ~2s to ~0.5s.
2. Parallelize a bit more of the helper tests.
2022-08-02 17:58:59 +03:00