[
https://issues.apache.org/jira/browse/HDDS-17843?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
rajesh updated HDDS-17843:
--------------------------
Description:
{{OzoneManagerRatisServer}} keeps and exposes the concrete
{{OzoneManagerStateMachine}}, and callers depend on the class instead of an
abstraction. Same kind of refactor as HDDS-17842 for SCM.
*Problems*
# {{OzoneManagerRatisServer.getOmStateMachine()}} returns the concrete class.
** Callers in {{OzoneManager}} use class-only methods: {{awaitApplied(...)}},
{{pause()}}, {{unpause(...)}} and {{awaitDoubleBufferFlush()}}.
** Callers cannot use the Ratis {{StateMachine}} interface, and the state
machine cannot be wrapped (for example by a tracing proxy) or substituted
without a raw-instance workaround.
# The state machine is built with a constructor flag, {{new
OzoneManagerStateMachine(this, TracingUtil.isTracingEnabled(conf))}}, which
only switches on tracing for the double buffer. A cross-cutting concern is
passed through the constructor.
# OM lifecycle operations are driven through the state machine: pause and
unpause around a DB checkpoint install, and waiting for the double buffer
flush. These are OM operations and not Ratis state machine behaviour.
*Proposal (to be refined)*
* Define a narrow interface, extending Ratis {{StateMachine}}, for what OM
needs from it (await applied index, pause and unpause, double buffer flush).
Make {{OzoneManagerRatisServer}} and {{OzoneManager}} depend on it.
* Look at moving pause, unpause and the double buffer flush behind an OM-owned
component, so the state machine only handles Ratis concerns.
* Create the state machine through {{TracingUtil.createInternalProxy}} so every
Ratis call can be traced without manual spans, and drop the
{{isTracingEnabled}} constructor flag if nothing else needs it.
*Benefits*
* Easy proxying and decoration, with no raw-instance workaround.
* OM code can be tested against the interface without a full state machine.
* The state machine shrinks to Ratis concerns, and tracing no longer has to be
passed through the constructor.
*Out of scope:* any behaviour change, and any change to the Ratis log or the OM
protocol.
was:
{{OzoneManagerRatisServer}} keeps and exposes the concrete
{{OzoneManagerStateMachine}}, and callers depend on the class instead of an
abstraction. Same kind of refactor as HDDS-17842 for SCM.
_Problems_
h1. {{OzoneManagerRatisServer.getOmStateMachine()}} returns the concrete class.
#* Callers in {{OzoneManager}} use class-only methods: {{awaitApplied(...)}},
{{pause()}}, {{unpause(...)}} and {{awaitDoubleBufferFlush()}}.
#* Callers cannot use the Ratis {{StateMachine}} interface, and the state
machine cannot be wrapped (for example by a tracing proxy) or substituted
without a raw-instance workaround.
h1. The state machine is built with a constructor flag, {{new
OzoneManagerStateMachine(this, TracingUtil.isTracingEnabled(conf))}}, which
only switches on tracing for the double buffer. A cross-cutting concern is
passed through the constructor.
h1. OM lifecycle operations are driven through the state machine: pause and
unpause around a DB checkpoint install, and waiting for the double buffer
flush. These are OM operations and not Ratis state machine behaviour.
_Proposal (to be refined)_
* Define a narrow interface, extending Ratis {{StateMachine}}, for what OM
needs from it (await applied index, pause and unpause, double buffer flush).
Make {{OzoneManagerRatisServer}} and {{OzoneManager}} depend on it.
* Look at moving pause, unpause and the double buffer flush behind an OM-owned
component, so the state machine only handles Ratis concerns.
* Create the state machine through {{TracingUtil.createInternalProxy}} so every
Ratis call can be traced without manual spans, and drop the
{{isTracingEnabled}} constructor flag if nothing else needs it.
_Benefits_
* Easy proxying and decoration, with no raw-instance workaround.
* OM code can be tested against the interface without a full state machine.
* The state machine shrinks to Ratis concerns, and tracing no longer has to be
passed through the constructor.
_Out of scope:_ any behaviour change, and any change to the Ratis log or the OM
protocol.
> Refactor OzoneManagerStateMachine: stop exposing the concrete class from
> OzoneManagerRatisServer
> ------------------------------------------------------------------------------------------------
>
> Key: HDDS-17843
> URL: https://issues.apache.org/jira/browse/HDDS-17843
> Project: Apache Ozone
> Issue Type: Task
> Components: Ozone Manager
> Reporter: rajesh
> Priority: Major
>
> {{OzoneManagerRatisServer}} keeps and exposes the concrete
> {{OzoneManagerStateMachine}}, and callers depend on the class instead of an
> abstraction. Same kind of refactor as HDDS-17842 for SCM.
> *Problems*
> # {{OzoneManagerRatisServer.getOmStateMachine()}} returns the concrete class.
> ** Callers in {{OzoneManager}} use class-only methods: {{awaitApplied(...)}},
> {{pause()}}, {{unpause(...)}} and {{awaitDoubleBufferFlush()}}.
> ** Callers cannot use the Ratis {{StateMachine}} interface, and the state
> machine cannot be wrapped (for example by a tracing proxy) or substituted
> without a raw-instance workaround.
> # The state machine is built with a constructor flag, {{new
> OzoneManagerStateMachine(this, TracingUtil.isTracingEnabled(conf))}}, which
> only switches on tracing for the double buffer. A cross-cutting concern is
> passed through the constructor.
> # OM lifecycle operations are driven through the state machine: pause and
> unpause around a DB checkpoint install, and waiting for the double buffer
> flush. These are OM operations and not Ratis state machine behaviour.
> *Proposal (to be refined)*
> * Define a narrow interface, extending Ratis {{StateMachine}}, for what OM
> needs from it (await applied index, pause and unpause, double buffer flush).
> Make {{OzoneManagerRatisServer}} and {{OzoneManager}} depend on it.
> * Look at moving pause, unpause and the double buffer flush behind an
> OM-owned component, so the state machine only handles Ratis concerns.
> * Create the state machine through {{TracingUtil.createInternalProxy}} so
> every Ratis call can be traced without manual spans, and drop the
> {{isTracingEnabled}} constructor flag if nothing else needs it.
> *Benefits*
> * Easy proxying and decoration, with no raw-instance workaround.
> * OM code can be tested against the interface without a full state machine.
> * The state machine shrinks to Ratis concerns, and tracing no longer has to
> be passed through the constructor.
> *Out of scope:* any behaviour change, and any change to the Ratis log or the
> OM protocol.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]