Class LogicalUnit

Constructor for LogicalUnit.

This needs to be overridden in derived classes in order to check op validity.

Parameters:

• wconfdcontext ((int, string)) - the identity of the logical unit to represent itself to wconfd when asking for resources; it is given as job id and livelock file.
• wconfd - the wconfd class to use; dependency injection to allow testability.

Overrides: object.__init__

Return the list of locks owned by the LU at a given level.

This method assumes that is field wconfdlocks is set correctly by mcpu.

Return a request to release the specified locks of the given level.

Correctly break up the group lock to do so.

Check syntactic validity for the opcode arguments.

This method is for doing a simple syntactic check and ensure validity of opcode parameters, without any cluster-related checks. While the same can be accomplished in ExpandNames and/or CheckPrereq, doing these separate is better because:

• ExpandNames is left as as purely a lock-related function
• CheckPrereq is run after we have acquired locks (and possible waited for them)

The function is allowed to change the self.op attribute so that later methods can no longer worry about missing parameters.

Expand names for this LU.

This method is called before starting to execute the opcode, and it should update all the parameters of the opcode to their canonical form (e.g. a short node name must be fully expanded after this method has successfully completed). This way locking, hooks, logging, etc. can work correctly.

LUs which implement this method must also populate the self.needed_locks member, as a dict with lock levels as keys, and a list of needed lock names as values. Rules:

• use an empty dict if you don't need any lock
• if you don't need any lock at a particular level omit that level (note that in this case DeclareLocks won't be called at all for that level)
• if you need locks at a level, but you can't calculate it in this function, initialise that level with an empty list and do further processing in LogicalUnit.DeclareLocks (see that function's docstring)
• don't put anything for the BGL level
• if you want all locks at a level use locking.ALL_SET as a value

If you need to share locks (rather than acquire them exclusively) at one level you can modify self.share_locks, setting a true value (usually 1) for that level. By default locks are not shared.

This function can also define a list of tasklets, which then will be executed in order instead of the usual LU-level CheckPrereq and Exec functions, if those are not defined by the LU.

Examples:

 # Acquire all nodes and one instance
 self.needed_locks = {
   locking.LEVEL_NODE: locking.ALL_SET,
   locking.LEVEL_INSTANCE: ['instance1.example.com'],
 }
 # Acquire just two nodes
 self.needed_locks = {
   locking.LEVEL_NODE: ['node1-uuid', 'node2-uuid'],
 }
 # Acquire no locks
 self.needed_locks = {} # No, you can't leave it to the default value None

Declare LU locking needs for a level

While most LUs can just declare their locking needs at ExpandNames time, sometimes there's the need to calculate some locks after having acquired the ones before. This function is called just before acquiring locks at a particular level, but after acquiring the ones at lower levels, and permits such calculations. It can be used to modify self.needed_locks, and by default it does nothing.

This function is only called if you have something already set in self.needed_locks for the level.

Parameters:

• level (member of ganeti.locking.LEVELS) - Locking level which is going to be locked

Check prerequisites for this LU.

This method should check that the prerequisites for the execution of this LU are fulfilled. It can do internode communication, but it should be idempotent - no cluster or system changes are allowed.

The method should raise errors.OpPrereqError in case something is not fulfilled. Its return value is ignored.

This method should also update all the parameters of the opcode to their canonical form if it hasn't been done by ExpandNames before.

Execute the LU.

This method should implement the actual work. It should raise errors.OpExecError for failures that are somewhat dealt with in code, or expected.

Prepare the LU to run again.

This method is called if the Exec failed for temporarily lacking resources. It is expected to change the state of the LU so that it can be tried again, and also change its locking policy to acquire more resources to have a better chance of suceeding in the retry.

Build hooks environment for this LU.

Returns: dict

Dictionary containing the environment that will be used for running the hooks for this LU. The keys of the dict must not be prefixed with "GANETI_"--that'll be added by the hooks runner. The hooks runner will extend the environment with additional variables. If no environment should be defined, an empty dictionary should be returned (not None).

Build list of nodes to run LU's hooks.

Returns: tuple; (list, list)

Tuple containing a list of node UUIDs on which the hook should run before the execution and a list of node UUIDs on which the hook should run after the execution. No nodes should be returned as an empty list (and not None).

Extend list of nodes to run the post LU hook.

This method allows LUs to change the list of node UUIDs on which the post hook should run after the LU has been executed but before the post hook is run.

Parameters:

• post_hook_node_uuids (list) - The initial list of node UUIDs to run the post hook on, as returned by BuildHooksNodes.

Returns: list

list of node UUIDs on which the post hook should run. The default implementation returns the passed in post_hook_node_uuids, but custom implementations can choose to alter the list.

Notify the LU about the results of its hooks.

This method is called every time a hooks phase is executed, and notifies the Logical Unit about the hooks' result. The LU can then use it to alter its result based on the hooks. By default the method does nothing and the previous result is passed back unchanged but any LU can define it if it wants to use the local cluster hook-scripts somehow.

Parameters:

• phase - one of constants.HOOKS_PHASE_POST or constants.HOOKS_PHASE_PRE; it denotes the hooks phase
• hook_results - the results of the multi-node hooks rpc call
• feedback_fn - function used send feedback back to the caller
• lu_result - the previous Exec result this LU had, or None in the PRE phase

Returns:

the new Exec result, based on the previous result and hook results

Called when the hooks get aborted by an exception.

This method is called everytime a hooks phase is aborted by an exception. This exception is most likely of type errors.HooksAbort. However, we keep the design of this function broad enough to handle any kind of exception.

The intended purpose of this call back is to run any action that is necessary to bring the cluster back to a clean state from the point in time before calling the hook.

Parameters:

• phase (string) - one of constants.HOOKS_PHASE_POST or constants.HOOKS_PHASE_PRE; it denotes the hooks phase
• feedback_fn (callable) - function used send feedback back to the caller
• exception (Exception) - The exception that was raised during the execution of hooks.