11 KiB
Pike Project Priorities
List of priorities the Watcher drivers team is prioritizing in Pike.
Cinder Model Intergration
Extend Watcher Cluster Data Model with Cinder-related data. There should be able following features: To integrate storage info at the model build stage
To consume all the needed Cinder Notifications in order to maintain the consistency of the storage-related part of the model
To easily query/retrieve the storage information from within a strategy via a clear set of methods
Audit tag in VM Metadata
When Watcher runs Audit to achieve a Goal, there should be some way for the application/VM owners to know that their VMs are under audit and is flagged for Action Plan execution. These information could be stored in VM metadata with a timestamp after which action plan will be executed.
Support Gnocchi in Watcher
Today, Watcher uses Telemetry and Monasca to collect metrics from the Cluster. There is need to support Gnocchi as well since Ceilometer v2 API is deprecated.
Noisy Neighbor Strategy
L3 cache is critical and limit system level resource shared by all apps or VMs on one node. If one VM occupies most of L3 cache, other VMs on the node likely starve without enough L3 cache thus poor performance.
This BP adds a new Strategy to detect then migrate such cache greedy VM based on some new cache/memory metrics.
Workload characterization grammar
As we run several workloads in cloud, we should be able to characterize such workloads as input to watcher for ensuring Application QoS, placements and consolidation.
An example of workload characterization is a weighted combination of CPU, Memory or any other resource attributes like High IOPs, Network latency etc.
Scope of this blueprint is to come up with a grammar structure for defining workload character.
Stale the Action Plan
When an audit is created and launched successfully, it generates a new Action Plan with status RECOMMENDED. If the Cluster Data Model has changed by and by, the action plan is still keep the RECOMMENDED state. There is not an expiry date or event that can invalidate the action plan by far.
Workload Characterization and QoS
Based on the defined workload characteristics we should be able to apply Quality of Services to applications. An example would be leveraging technologies like Intel RDT.
This opens up several application optimization possibilities (use cases like NFV etc.) and also ensures efficient use of cloud resources. Scope of this blueprint is to build a QoS strategy using Intel RDT and workload grammar.
Action versioned notifications
As of now, there is no way for any service (Watcher included) to know when an action has been created, modified or deleted. This prevents any form of event-based reaction which may be useful for 3rd party services or plugins.
This blueprint should define the list of Action notifications to be implemented as well as their respective payload structures.
Cancel Action Plan
As of now Administrator can update the action plan state to CANCELLED but there is no action taken by Watcher to cancel the action plan. It only updates the action plan state to CANCELLED.
It should be possible to CANCEL execution of the action plan by Watcher.
Dynamic Action Description
By introducing a new way, for developer, to implement a strategy with new customized actions (blueprint watcher-add-actions-via-conf), we have no more the possibility to have the literal description of an planned Action before to execute it (in the Watcher Applier). This literal description is important when the cloud admin want to see details information about a recommended action plan.
Power On and Power Off in Watcher
Watcher need one strategy which can reduce the power consumption.
A traffic system could be running on many virtual machines. The traffic is busy during day time, so the traffic system would increase virtual machines' number to satisfy its workload. But during the night, the traffic's workload decreases obviously, so this traffic system would delete redundant virtual machines. This feature we call "elastic scaling" in telecom.
The telecom operators have their own hardware equipment and sometimes the size of hardware is large. So telecom operators want to use cloud center manager software to reduce the energy consumption of hardware automatically based on "elastic scaling".
Suspended audit state
As of now Watcher have to delete audit and recreate audit if administrator want to stop creating action plan of audit with continuous mode.
This blueprint adds suspended audit state for stopping creation of action plan related to audit with continuous mode.
JSONschema validation
As of now in Watcher both jsonschema and voluptuous are used to validate JSON payloads. However, the problem with voluptuous is that its structure is not standardized compare to jsonschema which means that we cannot easily expose the validation schema through our API.
Service versioned notifications
As of now, there is no way for any service (Watcher included) to know when an action has been created, modified or deleted. This prevents any form of event-based reaction which may be useful for 3rd party services or plugins.
This blueprint should define the list of Service notifications to be implemented as well as their respective payload structures.
Notifications for action plan cancel
Notifications needs to be added to action and actionplan for new operation actionplan cancel.
Use cron syntax for CONTINUOUS audits
As of now we use a period in seconds to schedule continuous audits. This works well but does not really give the flexibility that an operator might actually want. Therefore, we should also provide a way to express out scheduling needs via the cron syntax which shall give operators a fine grained control.
This change implies the refactoring of the API so backward compatibility should be guaranteed. On the Watcher dashboard side, we should also provide an easy-to-use form to fill in this cron field.
We should also keep the cron syntax and the creation timestamp in the DB
Event-driven optimization based
We propose an event-driven optimization-based audit control. We wants to select among a list of events which may trigger the audit :
- React to a predicted situation.
- React to a critical situations and changes in system (e.g: threshold )
- A new compute node has been added to the cluster
- A compute node has been removed from the cluster
- A new virtual machine has been created