[Starlingx-discuss] Improve FM's Python API return data

Thu Mar 21 13:16:14 UTC 2019

Hi Shuicheng,

First of all, thank you for supporting this change for us. It's much appreciated.

Yes, please go with (what was in your spreadsheet)

True: alarm is cleared
False: alarm is not found.
Exception: there is operation failure.

Please apply that change to the other new APIs as well.
Looking at the code now more closely we need to ensure that the 'clear_fault', 'get_fault','clear_all' , 'get_faults' and 'get_faults_by_id' all follow this approach.

Interestingly I don't see the get_faults_by_id returning an exception in my code in the case where there are none found. It is behaving properly but looking at the code I can only assume that fm_core.get_by_aid is returning None rather than False which would explain it.

Please test 'each' the APIs with the following conditions ...

*         With applicable alarm present or success path

*         With applicable alarm not present.

*         With FM main process not running (you can do this by runnin the operations on a compute while there is no controller up.)

*         With an (software) error API call (like using an invalid alarmID)

Eric.

From: Lin, Shuicheng [mailto:shuicheng.lin at intel.com]
Sent: Wednesday, March 20, 2019 9:33 PM
To: MacDonald, Eric; Liu, Tao; Wensley, Barton; Bailey, Henry Albert (Al)
Cc: Xie, Cindy; starlingx-discuss at lists.starlingx.io; Khalil, Ghada
Subject: RE: Improve FM's Python API return data
Importance: High

Hi Eric,
Thanks for using the new API. :)
I will update code to take "clear non exist alarm" as normal case (no exception) as your suggestion.

Another thing is, I  don't think we need return value for clear_fault/clear_all API.
If we want return value for clear_fault/clear_all, then we could define return value as below:
True: alarm is cleared
False: alarm is not found.
Exception: there is operation failure.
Which version do you prefer?
Thanks.

I created below LP issue to track it:
https://bugs.launchpad.net/starlingx/+bug/1821112

Best Regards
Shuicheng

From: MacDonald, Eric [mailto:Eric.MacDonald at windriver.com]
Sent: Thursday, March 21, 2019 2:10 AM
To: Lin, Shuicheng <shuicheng.lin at intel.com>; Liu, Tao <Tao.Liu at windriver.com>; Wensley, Barton <Barton.Wensley at windriver.com>; Bailey, Henry Albert (Al) <Al.Bailey at windriver.com>
Cc: Xie, Cindy <cindy.xie at intel.com>; starlingx-discuss at lists.starlingx.io; Khalil, Ghada <Ghada.Khalil at windriver.com>
Subject: RE: Improve FM's Python API return data

Hi Shuicheng,

I've been using your new API FaultAPIsV2 and found that it is not behaving the way we discussed or I expected.
I looked back on the code update and this behavior seems to have gotten by all of us.

My example is for the clear_fault case.

I call clear_fault against an alarmID/entity path that there is no alarm raised for but instead of just returning False or None ; it generates an exception. From below , from Bart who I agree with ...

"My view is that you should be updating all six APIs to be consistent and raise an exception if the operation fails (e.g. because the API can't connect to fmmanager)."

... I expected to get an exception pretty much only if there was a 'software error' in the API call or it was 'unable to connect to the fm manager'. Not simply if the referenced alarm does not exist.

>From your SS my case is the highlighted

[cid:image001.jpg at 01D4DFC0.AC313D50]

Eric.

From: Lin, Shuicheng [mailto:shuicheng.lin at intel.com]
Sent: Tuesday, February 26, 2019 7:47 PM
To: Liu, Tao; Wensley, Barton; MacDonald, Eric; Bailey, Henry Albert (Al)
Cc: Xie, Cindy; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data
Importance: High

Hi all,
Thanks for the quick confirmation. I will implement the code as Bart's suggestion.

Best Regards
Shuicheng

From: Liu, Tao [mailto:Tao.Liu at windriver.com]
Sent: Tuesday, February 26, 2019 10:31 PM
To: Wensley, Barton <Barton.Wensley at windriver.com<mailto:Barton.Wensley at windriver.com>>; Lin, Shuicheng <shuicheng.lin at intel.com<mailto:shuicheng.lin at intel.com>>; MacDonald, Eric <Eric.MacDonald at windriver.com<mailto:Eric.MacDonald at windriver.com>>; Bailey, Henry Albert (Al) <Al.Bailey at windriver.com<mailto:Al.Bailey at windriver.com>>
Cc: Xie, Cindy <cindy.xie at intel.com<mailto:cindy.xie at intel.com>>; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data

Hi Shuicheng,
I agree with Bart's recommendation.
The failure condition occurs when the FM manager has not been started by SM during the booting or swact, and raise an exception for this condition is a right approach.
Having a new class for the new behavior is a good way to migrate the API changes.

Regards,
Tao

From: Wensley, Barton
Sent: Tuesday, February 26, 2019 9:11 AM
To: Lin, Shuicheng; MacDonald, Eric; Liu, Tao; Bailey, Henry Albert (Al)
Cc: Xie, Cindy; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data

Shuicheng,

As per your spreadsheet, there are actually three APIs that currently return ambiguous results (get_fault, get_faults and get_faults_by_id). My view is that you should be updating all six APIs to be consistent and raise an exception if the operation fails (e.g. because the API can't connect to fmmanager).

You can avoid impacting the existing APIs by having both a FaultAPIs class and a FaultAPIsV2 class. To avoid code duplication, you would create a FaultAPIsBase class to hold all the common code and subclass from it.

Bart

From: Lin, Shuicheng [mailto:shuicheng.lin at intel.com]
Sent: February 25, 2019 8:59 PM
To: MacDonald, Eric; Wensley, Barton; Liu, Tao; Bailey, Henry Albert (Al)
Cc: Xie, Cindy; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data

Hi Eric/Tao/Al,
Please confirm you agree a new set fault API (FaultAPIs_V2) or not.
The attached excel is a simple comparison between original class FaultAPIs and FaultAPIs_V2.
For myself, I still prefer to add 2 new API in original class, so other API call could be kept without change.
We could have more discuss about the FaultAPIs_V2 design if you all agree to choose it.

Best Regards
Shuicheng

From: MacDonald, Eric [mailto:Eric.MacDonald at windriver.com]
Sent: Monday, February 25, 2019 10:23 PM
To: Wensley, Barton <Barton.Wensley at windriver.com<mailto:Barton.Wensley at windriver.com>>; Lin, Shuicheng <shuicheng.lin at intel.com<mailto:shuicheng.lin at intel.com>>; Liu, Tao <Tao.Liu at windriver.com<mailto:Tao.Liu at windriver.com>>; Bailey, Henry Albert (Al) <Al.Bailey at windriver.com<mailto:Al.Bailey at windriver.com>>
Cc: Xie, Cindy <cindy.xie at intel.com<mailto:cindy.xie at intel.com>>; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data

I see your point Bart.
This Jira is an Enhancement Request and fulfilling that should not be done so half way.
All the APIs in V2 should be changed to defer exception handling to the caller as apparently that is the pythonic way.

From: Wensley, Barton
Sent: Monday, February 25, 2019 8:10 AM
To: Lin, Shuicheng; MacDonald, Eric; Liu, Tao; Bailey, Henry Albert (Al)
Cc: Xie, Cindy; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data
Importance: High

I disagree with this approach. Instead of just updating two of the methods, you should be updating all the methods in the FaultAPIs class that could fail in the same way and have them raise an exception. You should create a new FaultAPIs_v2 class for the updated methods to avoid impacting existing users. When a user of this API wants to use the new exceptions, they will start using the new class.

Bart

From: Lin, Shuicheng [mailto:shuicheng.lin at intel.com]
Sent: February 25, 2019 2:58 AM
To: MacDonald, Eric; Liu, Tao; Bailey, Henry Albert (Al); Wensley, Barton
Cc: Xie, Cindy; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: RE: Improve FM's Python API return data

Hi all,
Based on the comments in patch [0], I will add two new API, get_fault_ex and get_faults_by_id_ex.
I will not add a new class, since all other APIs are kept without change.
For get_fault_ex and get_faults_by_id_ex:
the return value will be alarm list (maybe None).
Exception will be thrown if there is error during API execution.

[0]: https://review.openstack.org/637655

Best Regards
Shuicheng

From: Lin, Shuicheng [mailto:shuicheng.lin at intel.com]
Sent: Wednesday, February 20, 2019 10:41 AM
To: MacDonald, Eric <Eric.MacDonald at windriver.com<mailto:Eric.MacDonald at windriver.com>>; Liu, Tao <Tao.Liu at windriver.com<mailto:Tao.Liu at windriver.com>>; Al.Bailey at windriver.com<mailto:Al.Bailey at windriver.com>
Cc: Xie, Cindy <cindy.xie at intel.com<mailto:cindy.xie at intel.com>>; starlingx-discuss at lists.starlingx.io<mailto:starlingx-discuss at lists.starlingx.io>
Subject: Re: [Starlingx-discuss] Improve FM's Python API return data

Hi Eric,
Thanks for the detail comments.
In my plan, the return value for get_fault API will be like below:
(False, None): there is error in the API execution.
(True, None): API execution correctly, and there is no alarm message.
(True, Alarm): API execution correctly, and there is alarm message.

I think it should fix the issue you have. Is it right?
If you agree on this new API return value change, I will implement it.
Thanks.

Best Regards
Shuicheng

From: MacDonald, Eric [mailto:Eric.MacDonald at windriver.com]
Sent: Tuesday, February 19, 2019 10:35 PM
To: Lin, Shuicheng <shuicheng.lin at intel.com<mailto:shuicheng.lin at intel.com>>; Liu, Tao <Tao.Liu at windriver.com<mailto:Tao.Liu at windriver.com>>
Cc: Xie, Cindy <cindy.xie at intel.com<mailto:cindy.xie at intel.com>>
Subject: RE: Improve FM's Python API return data

I took a look and added comments. I don't think this update fixes the issue I have.

From: Lin, Shuicheng [mailto:shuicheng.lin at intel.com]
Sent: Monday, February 18, 2019 10:09 PM
To: MacDonald, Eric; Liu, Tao
Cc: Xie, Cindy
Subject: RE: Improve FM's Python API return data
Importance: High

Resend the mail.

Best Regards
Shuicheng

From: Lin, Shuicheng
Sent: Tuesday, February 19, 2019 9:56 AM
To: Liu, Tao (Wind River) <tao.liu at windriver.com<mailto:tao.liu at windriver.com>>; 'eric.macdonald at windriver.com' <eric.macdonald at windriver.com<mailto:eric.macdonald at windriver.com>>
Cc: Xie, Cindy <cindy.xie at intel.com<mailto:cindy.xie at intel.com>>
Subject: Improve FM's Python API return data

Hi Tao/Eric,
For story 2004859<https://storyboard.openstack.org/#!/story/2004859>, I plan to modify the get_fault api return value as a tuple.
(True/False, Alarm)
Alarm is valid only when the 1st item is True.

Sample code is uploaded for you early review:
Stx-fault: https://review.openstack.org/637655
Stx-integ: https://review.openstack.org/637656
Could you help review it and share your thought?
If you agree on it, I will implement the code and test it.

For test, I will try to kill fmManager/modify its listening port to simulate the failure case.
For the api call outside of stx-fault, I may try to call it manually to verify.
Please share me if there is better verify method.
Thanks.

Best Regards
Shuicheng

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.starlingx.io/pipermail/starlingx-discuss/attachments/20190321/1b4bc264/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.jpg
Type: image/jpeg
Size: 22912 bytes
Desc: image001.jpg
URL: <http://lists.starlingx.io/pipermail/starlingx-discuss/attachments/20190321/1b4bc264/attachment-0001.jpg>