Upgrading

Pull the new image and restart. The database upgrades itself on the first start.

Most installations need to do nothing else. Each version below lists the setups that need an action. If none applies to you, you are done.

From 0.4 to 1.0

Your setupStep
Connectors declared in quanthea.yaml1
Elasticsearch or OpenSearch connectors declared in the file1, 2
Active alerts on Elasticsearch, OpenSearch or MongoDB3
SQL queries that call file, network or server functions4
HTTP connectors whose API redirects5
Stats or tables over an Elasticsearch or OpenSearch time series6

1. Connectors in the file hide system objects

A connector declared in quanthea.yaml now hides the source’s own system objects, and a query on one is refused. Connectors made in the app are not affected.

SourceHidden
Elasticsearch, OpenSearchIndices that start with ., security-auditlog-*, top_queries-*, ilm-history-*
PostgreSQL, TimescaleDBinformation_schema, pg_catalog, pg_* and TimescaleDB’s own schemas
MySQL, MariaDBinformation_schema, performance_schema, mysql, sys
ClickHousesystem, information_schema
Trino, InfluxDBinformation_schema, system
MongoDBsystem. collections

If a dashboard, alert or report reads one, add this to that connector and restart:

connectors:
  warehouse:
    kind: postgres
    visibility:
      showSystem: true

A panel that reads a hidden object shows an error that names it. See visibility.

2. Elasticsearch and OpenSearch: one more permission

A connector that hides indices resolves index names before each search. Grant its credentials:

  • Elasticsearch: the index privilege view_index_metadata.
  • OpenSearch: the action indices:admin/resolve/index.

Or set visibility.showSystem: true on the connector: it then hides and resolves nothing. Without either, every search on that connector fails with a permission error.

3. Alerts that use __interval on a search or MongoDB

An alert whose query names __interval (__interval_ms on MongoDB) is now checked on the same buckets its replay chart shows:

LookbackBucket
up to 2 hours1 minute
up to 12 hours5 minutes
up to 36 hours10 minutes
up to 4 days30 minutes
longer1 hour

Before, a check used much finer buckets, so a count per bucket is now larger. Open each such alert, look at its replay, and set its threshold again if it fires too often.

4. SQL functions that reach past the tables are refused

This applies to stored dashboards, alerts and reports too. A panel that calls one shows an error that names the function. Rewrite the query without it.

SourceRefused, among others
PostgreSQLpg_read_file, lo_export, dblink, set_config, pg_notify, query_to_xml
MySQLLOAD_FILE, sys_exec, GET_LOCK
ClickHouseurl, file, remote, s3, mysql, executable
TrinoTABLE(catalog.system.query(...))

5. HTTP connectors follow redirects only to allowed paths

If your API redirects to another path, add that path to the connector’s allowed paths. A path, or a variable value in a path, that holds a .. segment is refused.

6. Search time series return their empty buckets

On Elasticsearch and OpenSearch, a time histogram now returns every bucket of the time range: 0 for a count, no value for an average. A stat or a table that takes a mean or a minimum over it can show a different number. To keep the old number, add "min_doc_count": 1 to the query’s date_histogram and save a new version. Alerts are not affected.

After the upgrade

  1. Run the connection test of each connector declared in the file.
  2. Open one pinned dashboard per connector and check that its panels load.
  3. A panel error that names an object: step 1. Every search failing with a permission error: step 2.
  4. Look at the replay of each active alert on a search or MongoDB connector (step 3).