Elektra
0.8.19
|
You should first read elektra-plugins(7) to get an idea about plugins.
This document describes how elektra-plugins(7) are ordered with elektra-backends(7).
Multiple plugins open up many ways in which they can be arranged. A simple way is to have one array with pointers to plugins. To store a KeySet
, the first plugin starts off with the KeySet
passed to it. The resulting KeySet
is given to the next plugin. This is repeated for every plugin in the array.
To obtain a KeySet
, the array of plugins is processed the other way round. An empty KeySet
is passed to the last plugin. The resulting KeySet
is handed over to the previous one until the first plugin is executed.
This approach has shown not to be powerful enough to express all use cases by counter-evidence. Logging should take place after the storage plugin performs its actions in both directions. It is not possible to do this with a single array processed in a way as described above.
Two individual lists (get+set list) of plugins solve the described problem. Instead of bidirectional processing of a single list two separate arrays are used. It turns out that error scenarios also make this approach unsuccessful. When a plugin fails, no other plugin must be executed later on because it might depend on the previous plugin to work correctly. In kdbGet()
, this works well – the update process will be stopped. But during kdbSet()
, changes to the file system must be reverted. Plugins can leave a lock, temporary file or journal. These resources need to be cleaned up properly.
So every backend additionally needs a third array that is executed in error scenarios during kdbSet()
. Plugins responsible for the cleanup, rollback or error notification are inserted into it.
The resolver plugin requires this error list to do a proper rollback. Another use case is logging after a failure has happened.
The ordering of plugins inside these three arrays is controlled by elektra-contracts(7). Each of the three arrays has ten slots. These slots have names to be referred to in the contract.
Here you see a table that contains all names:
0 prerollback getresolver setresolver 1 prerollback pregetstorage presetstorage 2 prerollback pregetstorage presetstorage 3 prerollback pregetstorage presetstorage 4 prerollback pregetstorage presetstorage 5 rollback getstorage setstorage 6 postrollback postgetstorage precommit 7 postrollback postgetstorage commit 8 postrollback postgetstorage postcommit 9 postrollback postgetstorage postcommit
how the placement is influenced using infos/placement, infos/ordering and infos/stacking as described in CONTRACT.ini.