Normalise process state information for calculation nodes #1125
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
sphuber
force-pushed
the
fix_1062_normalise_calc_state
branch
11 times, most recently
from
f415854
to
05f2c83
Compare
The various Calculation types were using different styles of state information as well as the methods to inquire about it. For example JobCalculations have their own state in the CalcState, which was used in the old system to move the calculation through its states during execution, With the advent of the Process system, WorkCalculations were introduced, which did not have their own calculation state. Rather they use the state that the Process provides for them. However, users cannot always access the Process and so the state has to be exposed through the calculation node. This is now done by setting the PROCESS_STATE_KEY in the attributes of the calculation node when the process goes through a state change. With JobProcesses and the recent port of InlineCalculations to workfunctions, all calculations will now have this attribute set. This means we can defined the state information properties in the base AbstractCalculation class. Plum defines three active states: * CREATED * RUNNING * WAITING and three terminal states: * FINISHED * KILLED * EXCEPTED where FINISHED means the process finished nominally, EXCEPTED means the process incurred an unhandled exception and was handled by plum and KILLED means the process was requested to be killed from the outside, for example by the user. We map these process states on 6 meta state properties * terminated: The process is finished, killed or excepted * finished: The process is FINISHED, with successful or not result * killed: The process is KILLED * excepted: The process is EXCEPTED * finished_ok: The process is FINISHED and the execution was successful * failed: The process is FINISHED but the execution was not successful To distinguish between finished_ok and failed, the legacy JobCalculation already has a mechanism, namely the calculation state. All of the *FAILED states will indicate a failed calculation, whereas only CalcState.FINISHED means it was successful. The Process, and therewith WorkChain, does not have a mechanism to set an exit status yet, but that will be handled in another issue. When that is provided it should be used in the finished_ok and failed methods to distinguish between those two variants of ProcessState.FINISHED
The previous requirement of unicode was to strict. Changing to basestring will cover all acceptable cases
This will test that both style of launching JobCalculations (either the old way of directly creating a JobCalculation instance, storing it and calling submit on it, as well as the new way of directly creating a JobProcess) is supported by the new daemon.
Having 'is_finished' instead of 'finished' is more clear that it will return a boolean. Therefore all state quering properties of the Calculation class now have that prefix * is_terminated * is_finished * is_excepted * is_killed * is_finished_ok * is_failed Also fixed tests involving legacy workflows. For testing purposed these dummy workflows attached job calculations to already set the state to FINISHED and were being taken submitted or taken through the steps. As a result, the machinery checking whether they had finished, which now relies on the process state being set, was never finishing and the tests would run indefinitely
sphuber
force-pushed
the
fix_1062_normalise_calc_state
branch
14 times, most recently
from
8b999dd
to
3a3744f
Compare
For some as of yet unknown reason, having the interval for ssh at 30 will fail the daemon tests on Travis, but only for SqlA. The last job calculation that gets submitted is properly finished, just like all the other ones, except that its process state attribute is not updated to be Finished, but rather is stuck in Waiting. This means that the conditional for all calculations being Terminated is not satisified (even though all job calculation are FINISHED) and the test simply times out.
sphuber
force-pushed
the
fix_1062_normalise_calc_state
branch
from
3a3744f
to
0e3c7b5
Compare
muhrin
suggested changes
Feb 21, 2018
| return ProcessState(state) | ||
|
|
||
| @process_state.setter | ||
| def process_state(self, state): |
muhrin
approved these changes
Feb 21, 2018
This was referenced Feb 21, 2018
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #1062
The various Calculation types were using different styles of state information
as well as the methods to inquire about it. For example JobCalculations have
their own state in the CalcState, which was used in the old system to move the
calculation through its states during execution, With the advent of the Process
system, WorkCalculations were introduced, which did not have their own calculation
state. Rather they use the state that the Process provides for them. However,
users cannot always access the Process and so the state has to be exposed through
the calculation node. This is now done by setting the PROCESS_STATE_KEY in the
attributes of the calculation node when the process goes through a state change.
With JobProcesses and the recent port of InlineCalculations to workfunctions, all
calculations will now have this attribute set. This means we can defined the state
information properties in the base AbstractCalculation class. Plum defines three
active states:
and three terminal states:
where FINISHED means the process finished nominally, EXCEPTED means the process
incurred an unhandled exception and was handled by plum and KILLED means the process
was requested to be killed from the outside, for example by the user.
We map these process states on 6 meta state properties
To distinguish between finished_ok and failed, the legacy JobCalculation already
has a mechanism, namely the calculation state. All of the *FAILED states will
indicate a failed calculation, whereas only CalcState.FINISHED means it was
successful.
The Process, and therewith WorkChain, does not have a mechanism to set an exit
status yet, but that will be handled in another issue. When that is provided
it should be used in the finished_ok and failed methods to distinguish between
those two variants of ProcessState.FINISHED