FAQ and troubleshooting

Large migrations blocked

Tiger Cloud automatically manages the underlying disk volume. Due to platform limitations, it is only possible to resize the disk once every six hours. Depending on the rate at which you're able to copy data, you may be affected by this restriction. Affected instances are unable to accept new data and error with: FATAL: terminating connection due to administrator command.

If you intend on migrating more than 400 GB of data to Tiger Cloud, open a support request requesting the required storage to be pre-allocated in your Tiger Cloud service.

You can open a support request directly from Tiger Console, or by email to support@tigerdata.com.

Dumping and locks

When pg_dump starts, it takes an ACCESS SHARE lock on all tables which it dumps. This ensures that tables aren't dropped before pg_dump is able to drop them. A side effect of this is that any query which tries to take an ACCESS EXCLUSIVE lock on a table is blocked by the ACCESS SHARE lock.

A number of Tiger Cloud-internal processes require taking ACCESS EXCLUSIVE locks to ensure consistency of the data. The following is a non-exhaustive list of potentially affected operations:

• converting a chunk into the columnstore/rowstore and back
• continuous aggregate refresh (before 2.12)
• create hypertable with foreign keys, truncate hypertable
• enable hypercore on a hypertable
• drop chunks

The most likely impact of the above is that background jobs for retention policies, columnstore compression policies, and continuous aggregate refresh policies are blocked for the duration of the pg_dump command. This may have unintended consequences for your database performance.

Dumping with concurrency

When using the pg_dump directory format, it is possible to use concurrency to use multiple connections to the source database to dump data. This speeds up the dump process. Due to the fact that there are multiple connections, it is possible for pg_dump to end up in a deadlock situation. When it detects a deadlock it aborts the dump.

In principle, any query which takes an ACCESS EXCLUSIVE lock on a table causes such a deadlock. As mentioned above, some common operations which take an ACCESS EXCLUSIVE lock are:

• retention policies
• columnstore compression policies
• continuous aggregate refresh policies

If you would like to use concurrency nonetheless, turn off all background jobs in the source database before running pg_dump, and turn them on once the dump is complete. If the dump procedure takes longer than the continuous aggregate refresh policy's window, you must manually refresh the continuous aggregate in the correct time range. For more information, consult the refresh policies documentation.

To turn off the jobs:

SELECT public.alter_job(id::integer, scheduled=>false)
FROM _timescaledb_config.bgw_job
WHERE id >= 1000;

To turn on the jobs:

SELECT public.alter_job(id::integer, scheduled=>true)
FROM _timescaledb_config.bgw_job
WHERE id >= 1000;

Restoring with concurrency

If the directory format is used for pg_dump and pg_restore, concurrency can be employed to speed up the process. Unfortunately, loading the tables in the timescaledb_catalog schema concurrently causes errors. Furthermore, the tsdbadmin user does not have sufficient privileges to turn off triggers in this schema. To get around this limitation, load this schema serially, and then load the rest of the database concurrently.

# first, serially load JUST the _timescaledb_catalog
pg_restore -d "$TARGET" \
    --format=directory \
    --schema='_timescaledb_catalog' \
    --exit-on-error \
    --no-tablespaces \
    --no-owner \
    --no-privileges \
    dump

# next, concurrently load everything EXCEPT the _timescaledb_catalog
pg_restore -d "$TARGET" \
    --format=directory \
    --jobs=8 \
    --exclude-schema='_timescaledb_catalog' \
    --exit-on-error \
    --disable-triggers \
    --no-tablespaces \
    --no-owner \
    --no-privileges \
    dump

Ownership of background jobs

The _timescaledb_config.bgw_jobs table is used to manage background jobs. This includes custom jobs, columnstore compression policies, retention policies, and continuous aggregate refresh policies. On Tiger Cloud, this table has a trigger which ensures that no database user can create or modify jobs owned by another database user. This trigger can provide an obstacle for migrations.

If the --no-owner flag is used with pg_dump and pg_restore, all objects in the target database are owned by the user that ran pg_restore, likely tsdbadmin.

If all the background jobs in the source database were owned by a user of the same name as the user running the restore (again likely tsdbadmin), then loading the _timescaledb_config.bgw_jobs table should work.

If the background jobs in the source were owned by the postgres user, they are automatically changed to be owned by the tsdbadmin user. In this case, one just needs to verify that the jobs do not make use of privileges that the tsdbadmin user does not possess.

If background jobs are owned by one or more users other than the user employed in restoring, then there could be issues. To work around this issue, do not dump this table with pg_dump. Provide either --exclude-table-data='_timescaledb_config.bgw_job' or --exclude-table='_timescaledb_config.bgw_job' to pg_dump to skip this table. Then, use psql and the COPY command to dump and restore this table with modified values for the owner column.

# dump the _timescaledb_config.bgw_job table to a csv file replacing the owner
# values with tsdbadmin
psql -d "$SOURCE" -X -v ON_ERROR_STOP=1 --echo-errors -f - <<'EOF'
begin;
select string_agg
( case attname
    when 'owner' then $$'tsdbadmin' as "owner"$$
    else format('%I', attname)
  end
, ', '
) as cols
from pg_namespace n
inner join pg_class c
on (n.nspname = '_timescaledb_config'
and n.oid = c.relnamespace
and c.relname = 'bgw_job')
inner join pg_attribute a
on (c.oid = a.attrelid and a.attnum > 0)
\gset
copy
(
    select :cols
    from _timescaledb_config.bgw_job
    where id >= 1000
) to stdout with (format csv, header true)
\g bgw_job.csv
rollback;
\q
EOF

# copy the csv file into the target's _timescaledb_config.bgw_job
psql -X -d "$TARGET" -v ON_ERROR_STOP=1 --echo-errors \
    -c "\copy _timescaledb_config.bgw_job from 'bgw_job.csv' with (format csv, header match)"

Once the table has been loaded and the restore completed, you may then use SQL to adjust the ownership of the jobs and/or the associated stored procedures and functions as you wish.

Extension availability

There are a vast number of PostgreSQL extensions available in the wild. Tiger Cloud supports many of the most popular extensions, but not all extensions. Before migrating, check that the extensions you are using are supported on Tiger Cloud.

TimescaleDB extension in the public schema

When self-hosting, the TimescaleDB extension may be installed in an arbitrary schema. Tiger Cloud only supports installing the TimescaleDB extension in the public schema. How to go about resolving this depends heavily on the particular details of the source schema and the migration approach chosen.

Tablespaces

Tiger Cloud does not support using custom tablespaces. Providing the --no-tablespaces flag to pg_dump and pg_restore when dumping/restoring the schema results in all objects being in the default tablespace as desired.

Only one database per instance

While PostgreSQL clusters can contain many databases, Tiger Cloud services are limited to a single database. When migrating a cluster with multiple databases to Tiger Cloud, one can either migrate each source database to a separate Tiger Cloud service or "merge" source databases to target schemas.

Superuser privileges

The tsdbadmin database user is the most powerful available on Tiger Cloud, but it is not a true superuser. Review your application for use of superuser privileged operations and mitigate before migrating.

Migrate partial continuous aggregates

In order to improve the performance and compatibility of continuous aggregates, TimescaleDB v2.7 replaces partial continuous aggregates with finalized continuous aggregates.

To test your database for partial continuous aggregates, run the following query:

SELECT exists (SELECT 1 FROM timescaledb_information.continuous_aggregates WHERE NOT finalized);

If you have partial continuous aggregates in your database, upgrade them from partial to finalized before you migrate your database.

If you accidentally migrate partial continuous aggregates across PostgreSQL versions, you see the following error when you query any continuous aggregates:

ERROR:  insufficient data left in message.