Resuming services - Githubissues

golemfactory / yagna

An open platform and marketplace for distributed computations

GNU General Public License v3.0

399 stars 61 forks source link

Resuming services #1692

Closed nieznanysprawiciel closed 7 months ago

nieznanysprawiciel commented 2 years ago

Why:

Services must be able to start working after update without breaking Agreement

What:

[x] #1697

Tasks:

[x] #1778
[x] #1779
[x] https://github.com/golemfactory/ya-runtime-sdk/issues/20
[x] #1780
[ ] https://github.com/golemfactory/yapapi/issues/799 (SDK team)

mfranciszkiewicz commented 2 years ago

Edit: the former task list has been decomposed to a series of issues. The task list is not meant to exhaust the scope of changes needed for implementing the feature, but serve as a guidance during development

General remarks:

on each provider machine there are several yagna-ecosystem processes running simultaneously. In order to snapshot the computation state, we must allow for all components to shutdown gracefully; the components will shut down in a specific order
the system shutdown process must use SIGKILL as a last resort and try to rely on graceful component shutdown

maaktweluit commented 2 years ago

Nice overview!

One thing to add is that on yagna start-up the resuming of activities need to be triggered after payments have been initiated

jiivan commented 2 years ago

Nice overview!

One thing to add is that on yagna start-up the resuming of activities need to be triggered after payments have been initiated

So probably payment initiation should be permanent without the need to do this on every run.

nieznanysprawiciel commented 2 years ago

pause the following: output streaming

We should pause after command stop executing (either successfully or with timeout). Requestor still wants to get full stream

nieznanysprawiciel commented 2 years ago

resume activities, i.e. start ExeUnit(s) in previous working directories We need to restore whole Provider state for whole Agreement.

nieznanysprawiciel commented 2 years ago

@mfranciszkiewicz Can you propose and create tasks for this feature? This was part of the reasearch

mfranciszkiewicz commented 2 years ago

pause the following: output streaming

We should pause after command stop executing (either successfully or with timeout). Requestor still wants to get full stream

The next point states await current command termination (with a short timeout)

nieznanysprawiciel commented 2 years ago

pause the following: output streaming

We should pause after command stop executing (either successfully or with timeout). Requestor still wants to get full stream

The next point states await current command termination (with a short timeout)

I meant that order is wrong. First we need to wait and than pause