Add 'Troubleshooting' w/workaround for ownership

If a Postgres dump contains ALTER TABLE ... OWNER_TO <username>
statements which set the owner to a username different from
'synapse' the post Postgres import task will fail complaining
about lack of role.

Changing the matrix_postgres_connection_username group var has no
effect. However, the ALTER TABLE statements (and accompanying comments)
can be rewritten to change the username to 'synapse', which permits the
import task to succeed.

From a sample of 1, having the owner set in this was causes no
discernable side effects on the homeserver.
This commit is contained in:
bertiebaggio 2020-07-13 15:03:24 +01:00
parent c23a0620f2
commit 000b482d18

View file

@ -24,3 +24,54 @@ To import, run this command (make sure to replace `<server-path-to-postgres-dump
ansible-playbook -i inventory/hosts setup.yml --extra-vars='server_path_postgres_dump=<server-path-to-postgres-dump.sql>' --tags=import-postgres
**Note**: `<server-path-to-postgres-dump.sql>` must be a file path to a Postgres dump file on the server (not on your local machine!).
## Troubleshooting
A table ownership issue can occur if you are importing from a Synapse installation which was both:
- migrated from SQLite to Postgres, and
- used a username other than 'synapse'
In this case you may run into the following error during the import task:
```
"ERROR: role \"synapse_user\" does not exist"
```
where `synapse_user` is the database username from the previous Synapse installation.
This can be verified by examining the dump for ALTER TABLE statements which set OWNER TO that username:
```Shell
$ grep "ALTER TABLE" homeserver.sql"
ALTER TABLE public.access_tokens OWNER TO synapse_user;
ALTER TABLE public.account_data OWNER TO synapse_user;
ALTER TABLE public.account_data_max_stream_id OWNER TO synapse_user;
ALTER TABLE public.account_validity OWNER TO synapse_user;
ALTER TABLE public.application_services_state OWNER TO synapse_user;
...
```
It can be worked around by changing the username to `synapse`, for example by using `sed`:
```sed
$ sed -i "s/synapse_user/synapse/g" homeserver.sql"
```
This uses sed to perform an 'in-place' (`-i`) replacement globally (`/g`), searching for `synapse user` and replacing with `synapse` (`s/synapse_user/synapse`). If your database username was different, change `synapse_user` to that username instead.
Note that if the previous import failed with an error it may have made changes which are incompatible with re-running the import task right away; if you do so it may fail with an error such as:
```
ERROR: relation \"access_tokens\" already exists
```
In this case you can use the command suggested in the import task to clear the database before retrying the import:
```Shell
# systemctl stop matrix-postgres
# rm -rf /matrix/postgres/data/*
# systemctl start matrix-postgres
```
Once the database is clear and the ownership of the tables has been fixed in the SQL file, the import task should succeed.