diff options
Diffstat (limited to 'doc/04-Upgrading.md')
| -rw-r--r-- | doc/04-Upgrading.md | 106 |
1 files changed, 99 insertions, 7 deletions
diff --git a/doc/04-Upgrading.md b/doc/04-Upgrading.md index 5a085e0..74382e5 100644 --- a/doc/04-Upgrading.md +++ b/doc/04-Upgrading.md @@ -3,6 +3,98 @@ Specific version upgrades are described below. Please note that version upgrades are incremental. If you are upgrading across multiple versions, make sure to follow the steps for each of them. +## Upgrading to Icinga DB v1.2.0 + +Please apply the `1.2.0.sql` upgrade script to your database. For package installations, you can find this file at +`/usr/share/icingadb/schema/mysql/upgrades/` or `/usr/share/icingadb/schema/pgsql/upgrades/`, depending on your +database vendor. + +As the daemon checks the schema version, the recommended way to perform the upgrade is to stop the daemon, apply the +schema upgrade and then start the new daemon version. If you want to minimize downtime as much as possible, it is safe +to apply this schema upgrade while the Icinga DB v1.1.1 daemon is still running and then restart the daemon with the +new version. Please keep in mind that depending on the distribution, your package manager may automatically attempt to +restart the daemon when upgrading the package. + +!!! warning + + With MySQL and MariaDB, a locking issue can occur if the schema upgrade is applied while the history view is + accessed in Icinga DB Web. This can result in the upgrade being delayed unnecessarily and blocking other queries. + Please see [unblock history tables](#unblock-history-tables) for how to detect and resolve this situation. + +### Upgrading the state_history Table + +This release includes fixes for hosts and services reaching check attempt 256. However, on existing installations, +the schema upgrade required to fix the history tables isn't automatically applied by `1.2.0.sql` as a rewrite of the +whole `state_history` table is required. This can take a lot of time depending on the history size and the performance +of the database. During this time that table will be locked exclusively and can't be accessed otherwise. This means that +the existing history can't be viewed in Icinga Web and new history entries will be buffered in Redis®. + +There is a separate upgrade script `optional/1.2.0-history.sql` to perform the rewrite of the `state_history` table. +This allows you to postpone part of the upgrade to a longer maintenance window in the future, or skip it entirely +if you deem this safe for your installation. + +!!! warning + + Until `optional/1.2.0-history.sql` is applied, you'll have to lower `max_check_attempts` to 255 or less, otherwise + Icinga DB will crash with a database error if hosts/services reach check attempt 256. If you need to lower + `max_check_attempts` but want to keep the same timespan from an outage to a hard state, you can raise + `retry_interval` instead so that `max_check_attempts * retry_interval` stays the same. + +If you apply it, be sure that `1.2.0.sql` was already applied before. Do not interrupt it! At best use tmux/screen not +to lose your SSH session. + +### Unblock History Tables + +!!! info + + You don't need to read this section if you are using PostgreSQL. This applies to MySQL/MariaDB users only. + +In order to fix a loading performance issue of the history view in Icinga DB Web, this upgrade script adds an +appropriate index on the `history` table. Creating this new index normally takes place without blocking any other +queries. However, this may hang for a relatively considerable time, blocking all Icinga DB queries on all`*_history` +tables and the `history` table inclusively if there is an ongoing, long-running query on the `history` table. One way +of causing this to happen is if an Icinga Web user accesses the `icingadb/history` view just before you are running +this script. Depending on how many entries you have in the history table, Icinga DB Web may take quite a long time to +load, until your web servers timeout (if any) kicks in. + +When you observe that the upgrade script has been taking unusually long (`> 60s`) to complete, you can perform the +following analysis on another console and unblock it if necessary. It is important to note though that the script may +need some time to perform the reindexing on the `history` table even if it is not blocked. Nonetheless, you can use the +`show processlist` command to determine whether an excessive number of queries have been stuck in a waiting state. + +``` +MariaDB [icingadb]> show processlist; ++------+-----+-----+----------+-----+------+---------------------------------+------------------------------------+-----+ +| Id | ... | ... | db | ... | Time | State | Info | ... | ++------+-----+-----+----------+-----+------+---------------------------------+------------------------------------+-----+ +| 1475 | ... | ... | icingadb | ... | 1222 | Waiting for table metadata lock | INSERT INTO "notification_history" | ... | +| 1485 | ... | ... | icingadb | ... | 1262 | Creating sort index | SELECT history.id, history.... | ... | +| 1494 | ... | ... | icingadb | ... | 1224 | Waiting for table metadata lock | ALTER TABLE history ADD INDEX ... | ... | +| 1499 | ... | ... | icingadb | ... | 1215 | Waiting for table metadata lock | INSERT INTO "notification_history" | ... | +| 1500 | ... | ... | icingadb | ... | 1215 | Waiting for table metadata lock | INSERT INTO "state_history" ... | ... | +| ... | ... | ... | ... | ... | ... | ... | ... | ... | ++------+-----+-----+----------+-----+------+---------------------------------+------------------------------------+-----+ +``` + +In the above output are way too many Icinga DB queries, including the `ALTER TABLE history ADD INDEX` query from the +upgrade script, waiting for a metadata lock, they are just minimised to the bare essentials. Unfortunately, only one of +these queries is holding the `table metadata lock` that everyone else is now waiting for, which in this case is a +`SELECT` statement initiated by Icinga DB Web in the `icingadb/history` view, which takes an unimaginably long time. +Note that there might be multiple `SELECT` statements started before the upgrade script in your case when the history +view of your Icinga DB Web is opened by different Icinga Web users at the same time. + +You can now either just wait for the `SELECT` statements to finish by themselves and let them block the upgrade script +and all Icinga DB queries on all `*_history` tables or forcibly terminate them and let the remaining queries do their +work. In this case, cancelling that one blocking `SELECT` query will let the upgrade script continue normally without +blocking any other queries. +``` +MariaDB [icingadb]> kill 1485; +``` +In case you are insecure about which Icinga DB Web queries are blocking, you may simply cancel all long-running +`SELECT` statements listed with `show processlist` (see column `Time`). Cancelling a `SELECT` query will neither +crash Icinga DB nor corrupt your database, so feel free to abort every single one of them matching the Icinga DB +database (see column `db`). + ## Upgrading to Icinga DB v1.1.1 Please apply the `1.1.1.sql` upgrade script to your database. @@ -11,7 +103,7 @@ For package installations, you can find this file at `/usr/share/icingadb/schema Note that this upgrade will change the `history` table, which can take some time depending on the size of the table and the performance of the database. While the upgrade is running, that table will be locked and can't be accessed. This -means that the existing history can't be viewed in Icinga Web and new history entries will be buffered in Redis. +means that the existing history can't be viewed in Icinga Web and new history entries will be buffered in Redis®. As the daemon checks the schema version, the recommended way to perform the upgrade is to stop the daemon, apply the schema upgrade and then start the new daemon version. If you want to minimize downtime as much as possible, it is safe @@ -32,20 +124,20 @@ restart the daemon when upgrading the package. ## Upgrading to Icinga DB RC2 -Icinga DB RC2 is a complete rewrite compared to RC1. Because of this, a lot has changed in the Redis and database +Icinga DB RC2 is a complete rewrite compared to RC1. Because of this, a lot has changed in the Redis® and database schema, which is why they have to be deleted and recreated. The configuration file has changed from `icingadb.ini` to `config.yml`. Instead of the INI format, we are now using YAML and have introduced more configuration options. We -have also changed the packages of `icingadb-redis`, which is why the Redis CLI commands are now prefixed with `icingadb` -instead of just `icinga`, i.e. the Redis CLI is now accessed via `icingadb-redis-cli`. +have also changed the packages of `icingadb-redis`, which is why the Redis® CLI commands are now prefixed with `icingadb` +instead of just `icinga`, i.e. the Redis® CLI is now accessed via `icingadb-redis-cli`. Please follow the steps below to upgrade to Icinga DB RC2: 1. Stop Icinga 2 and Icinga DB. -2. Flush your Redis instances using `icinga-redis-cli flushall` (note the `icinga` prefix as we did not +2. Flush your Redis® instances using `icinga-redis-cli flushall` (note the `icinga` prefix as we did not upgrade `icingadb-redis` yet) and stop them afterwards. 3. Upgrade Icinga 2 to version 2.13.2 or newer. 4. Remove the `icinga-redis` package where installed as it may conflict with `icingadb-redis`. -5. Install Icinga DB Redis (`icingadb-redis`) on your primary Icinga 2 nodes to version 6.2.6 or newer. +5. Install Redis® (`icingadb-redis`) on your primary Icinga 2 nodes to version 6.2.6 or newer. 6. Upgrade Icinga DB to RC2. 7. Drop the Icinga DB MySQL database and recreate it using the provided schema. -8. Start Icinga DB Redis, Icinga 2 and Icinga DB. +8. Start Redis®, Icinga 2 and Icinga DB. |