There are four main operations in libelektra-kdb: open, get, set and close. For each of these there is a kdb* function the user calls to trigger the operation and plugins export a function for each of the operations they support (at least get).

Additionally, plugins may implement commit and error. These are part of the set operation and there is no corresponding kdbCommit or kdbError function available in libelektra-kdb.

The operations get and set also have different phases:

The get operation has: init, resolver, cachecheck, prestorage, storage and poststorage.
The set operation has: resolver, prestorage, storage and poststorage followed by precommit, commit and postcommit if the previous phases where successful or by prerollback, rollback and postrollback if the previous phases failed.

These phases are implemented by a backend plugin. Read the Documentation on Backend Plugins for more information on what backend plugins do.

Note The steps of the operations described below, are referenced in the source code with // Step X comments.

<tt>open</tt> Operation

The open operation implemented in kdbOpen is the first thing that happens to all KDB instances.

The basic flow of this operation is:

Create empty KDB instance
Configure KDB instance for bootstrap
Run bootstrap get operation: This loads the contents of system:/elektra/mountpoints so that the mountpoints can be configured.
Process contract and set up plugins for hooks (see Hooks)
Parse mountpoints: This transforms the contents of system:/elektra/mountpoints into the internal state stored in a KDB instance.
Reconfigure KDB with real mountpoints: This switches the KDB instance from bootstrap mode to use the real mountpoint state created above.
Add hard coded mountpoints to KDB instance: There are a few hard coded mountpoints (root mountpoints, system:/elektra/modules, system:/elektra/version, etc.) that are always present. They are added in this step.

Namespaces in mountpoint configs:

dir:/, user:/ and system:/ mountpoints can be created without restrictions, except for the reserved sections listed below.
spec:/ mountpoints can be created with the same restrictions, but they are also treated specially during get and set.
proc:/ mountpoints are always read-only and receive special treatment during get
default:/ mountpoints are read-only and receive special treatment during get, specifically they only go through the poststorage phase
mountpoints in all other namespaces are entirely illegal

Note The special treatments of the various namespaces are explained below in the sections for the get and set operation.

Reserved sections:

Creating a mountpoint for /elektra or below in any namespace is forbidden. This section of the KDB is reserved for Elektra's own config.
system:/elektra/mountpoints, user:/elektra/mountpoints and dir:/elektra/mountpoints are all required for the bootstrap process and use a hard coded backend. The backends are implemented by a standard file-based backend plugin that is defined at compile-time of libelektra-kdb.
system:/elektra/version and system:/elektra/modules will always use hard coded read-only backends containing information about this Elektra installation. The backends are implemented by special purpose backend plugins.

<tt>get</tt> Operation

The purpose of the get operation is to read data stored in backends into a KDB instance.

Note: Some details of a get operation are defined in the contract with backend plugins.

Properties of kdbGet():

After calling kdbGet (kdb, ks, parentKey), the KeySet ks will contain all keys (including their values) that are stored in any backend with a mountpoint that is below parentKey.
After calling kdbGet (kdb, ks, parentKey), below parentKey the KeySet ks will mostly contain keys that are stored in a backend. The exception here are proc:/ and spec:/ keys. For other namespaces, all keys below parentKey will be removed from ks. For proc:/ and spec:/ only keys that overlap with a backend that was loaded will be removed from ks.
The KeySet ks may contain other keys not below parentKey:
1. Keys that are not below parentKey, but are stored in a backend that contains other keys which are below parentKey. These keys are returned, because backends are treated as one atomic unit. Either all keys within a backend are read, or none of them are.
2. Keys that were already present in ks when kdbGet() was called and do not conflict with the goal of representing the current state of the KDB below parentKey.
After calling kdbGet (kdb, ks, parentKey), the Key parentKey will only have the meta:/error/* or meta:/warnings/#/* metakeys, if the errors/warnings originate from this kdbGet() call. In other words, kdbGet () first clears any existing errors/warnings and only then starts doing the actual work.
It is an error to use a parentKey with a namespace other than: default:/, proc:/, spec:, dir:/, user:/, system:/ or cascading

To the caller it looks as if kdbGet() had removed all keys below parentKey, as well as some others, from ks and then loaded the data from the backends. Which backends are actually read is an implementation detail. Which keys are removed from ks depends on the backends that are read.

kdbGet() will always try to be efficient in achieving its goal of reading the keys below parentKey. It is only guaranteed that below parentKey the KeySet ks correctly represents the state of the KDB. For the rest of ks there are no such guarantees.

Note: In the list below "phase" always refers to a phase of the get operation as described in the backend plugin contract.

The flow of this operation is:

Determine the backends needed to read all keys below parentKey
Run the open operation for all required backends that haven't been opened
Run the init phase on all the backends that haven't been initialized
Run the resolver phase on all backends
From now on ignore all backends, which indicated that there is no update.
If all backends are now ignored, return.
If a global cache plugin is enabled: Ask the global cache plugin for the cache handles (normally modification times) for all backends.
If all backends have an existing cache entry: Run the cachecheck phase on all backends
If all backends indicated the cache is still valid: Ask the global cache plugin for the cached data and return.
Run the prestorage and storage phase on all backends.
Run the poststorage phase of all spec:/ backends.
Merge the data from all backends
If enabled, run the gopts/get hook.
Run the spec/copy hook.
Split data back into individual backends.
Run the poststorage phase for all non-spec:/ backends.
Remove all keys which are below the parent key of any backend that has been read from ks.
Merge the data from all backends into ks.
If a global cache plugin is enabled, update cache.
Run the notification/send hook. Then return.

Note: In case of error, we abort immediately, restore ks to its original state and return.

Influence of namespaces:

spec:/ backends go through init, resolver, cache, presetstorage and storage phases as normal, but their poststorage phase is called earlier. This is required, because any validation and post-processing of spec:/ keys needs to happen, before they are used as the specification for other keys in the actual poststorage phase.
dir:/, user:/ and system:/ go through all phases as described above.
proc:/ mountpoints go through all the phases as described above, but they are not stored in the cache.
default:/ backends only go through the poststorage phase. This is because default:/ keys are generated from the specification (stored as spec:/ keys). Therefore, no default:/ keys can exist before the specification is processed by the spec/copy hook.
keys with other namespaces are always illegal in ks

<tt>set</tt> Operation

The purpose of the set operation is to write data from a KDB instance into backends.

Note: Some details of a set operation are defined in the contract with backend plugins.

Properties of kdbSet():

When calling kdbSet (kdb, ks, parentKey) the contents (key names, values and metadata) of ks will mostly not be modified. The only modifications that are made to ks are those that originate from the spec/copy hook.
All keys in ks that are below parentKey will be persisted in the KDB, when a kdbSet (kdb, ks, parentKey) call returns successfully. Additionally, any key in ks that shares a backend with another key which is below parentKey will also be persisted.
Calling kdbSet may result in an error, if kdbGet wasn't called on this KDB instance with the same parentKey at least once.
After calling kdbSet (kdb, ks, parentKey), the Key parentKey will only have the meta:/error/* or meta:/warnings/#/* metakeys, if the errors/warnings originate from this kdbSet() call. In other words, kdbSet () first clears any existing errors/warnings and only then starts doing the actual work.
It is an error to use a parentKey with a namespace other than: default:/, proc:/, spec:, dir:/, user:/, system:/ or cascading

The flow of this operation is:

Determine the backends needed to write all keys below parentKey.
Check that all backends are opened and initialized (i.e. kdbGet() was called).
Run the spec/copy hook on ks (to add metakeys to newly created keys).
Deep-Copy ks (below parentKey) into a new KeySet set_ks
Split set_ks into individual backends
Determine which backends contain changed data. Any backend that contains a key that needs sync (via KEY_FLAG_SYNC) could contain changed data. From now on ignore all backends that have not changed. From now on also ignore all backends that were initialized as read-only. Issue a warning, if a change was detected (via KEY_FLAG_SYNC) in a read-only backend.

Note: Steps 4-6 might be combined into a single procedure that deep-copies only keys from changed backends into separate KeySets per backend

7. Run the resolver and prestorage on all backends (abort immediately on error and go to e).
Merge the results into a new version of set_ks.
Run the spec/remove hook on set_ks (to remove copied metakeys).
Split set_ks into individual backends again.
Run the storage and poststorage phases on all backends (abort immediately on error and go to e).
If everything was successful: Run the precommit and commit phases on all backends (abort immediately on error and go to e), then run the postcommit phase on all backends (record all errors as warnings and ignore them) and return.

If there was an error: Run the prerollback, rollback and postrollback phases on all backends and return.

Influence of namespaces:

default:/ and proc:/ keys are completely ignored by kdbSet()
spec:/, dir:/, user:/ and system:/ go through all phases as described above.
keys with other namespaces are always illegal in ks

<tt>close</tt> Operation

The close operation is very simple. It simply frees up all resources used by a KDB instance.