Jump to: navigation, search

TaskFlow/Persistence

< TaskFlow
Revision as of 00:22, 14 October 2013 by Harlowja (talk | contribs) (Storage)

Revised on: 10/14/2013 by Harlowja

Overview

A persistence API as well as base persistence types are provided with taskflow for the purpose of ensuring that jobs, flows, and there associated tasks can be backed up in a database or in memory (or elsewhere). The user, when configuring the persistence API, has the option to specify which backend is desired and subsequently store and retrieve the data associated with the jobs, flows, and tasks in use.

Why?

Machovka harddisk.png
  • Allows for reconstruction and resumption of flows and there associated tasks.
  • Allows for redundant checks that expected data is provided.
  • Allows for the user to view the history of a jobs, flows and there associated tasks.
  • Facilitates debugging of taskflow usage and integration (and runtime/post-runtime analysis).

Design & Details

See: Details

Backends

Configuration

When configuring the backend to use, a stevedore driver (which uses python entrypoints) can be specified to locate the backend that your applications desires to use. This allows for easy extensibility of the backend that your application may plan to use (and does not limit the selection of backends to those that are included by default).

Defaults

  • SQLAlchemy:
    • Makes use of the sqlalchemy library to store all data in a (Postgres or MySQL or SQLite....) database.
    • Will be persisted in the event of a system failure.
  • In-memory:
    • Makes use of a in-memory dictionaries to store data in memory in a thread-safe manner.
    • Will NOT be persisted in the event of a system failure.
  • Local-files:
    • Makes use of a set of directories and files to store data in a thread and process safe manner.
    • Will be persisted in the event of a system failure.
  • More to come...

Storage

Now that we already have storage in taskflow -- that is the logbook (which is itself connected or derived/saved to a given backend). It should be emphasized that logbook is the authoritative, and, preferably, the only source of runtime state information. When task returns result, it should be written directly to logbook. When task or flow state changes in any way, logbook is first to know. Flow should not store task results -- there is logbook for that.

Logbook and a backend are responsible to store the actual data -- these together specify the persistence mechanism (how data is saved and where -- memory, database, whatever), and persistence policy (when data is saved -- every time it changes or at some particular moments or simply never). To make these components simpler to use we have come up with the concept of a storage API; this API allows engines to easily call into the storage layer and avoid the details about logbooks, flowdetails, taskdetails and backends.

Checkpointing

A WIP topic/discussion is the concept of check-pointing.

See: Checkpointing

Contributors

  • Kevin Chen (Rackspace)
  • Joshua Harlow (Yahoo!)
  • Jessica Lucci (Rackspace)
  • Ivan Melnikov (Grid Dynamics)