[ovirt-devel] improving VDSM error reporting

[Michal Skrivanek]

On May 13, 2014, at 19:00 , Maor Lipchuk <mlipchuk at redhat.com> wrote:

> The specific verb is not that relevant.
> it's already being done when using FAILED_HOT_SET_NUMBER_OF_CPUS,
> VDS_INITIALIZING, RUN_VM_FAILED or STORAGE_DOMAIN_ERROR (and more)
> 
> For general purposes we can use a new audit log such as this:
> RETUNED_VDSM_ERROR_MESSAGE=The error message returned by VDSM was:
> ${ErrorMessage}

Why do you want to invent a new message when there is one already? And rather put more useful info there
Pretty much as Dan suggests, though I'd be careful about direct passthrough of stack trace to UI.
I'd rather invest into review of these messages and adding the info we think can be useful, it may be libvirt's error code for some cases, but very often something else. The flows are complex and a particular message for one vdsm verb doesn't necessarily make sense on its own.

Thanks,
michal

> 
> Regards,
> Maor
> 
> On 05/13/2014 05:43 PM, Yoav Kleinberger wrote:
>> the specific example was CreateStorageDomain, not ConnectStorageDomain. 
>> but anyhow, the problem is a general one: add some more information to the errors that engine gets from VDSM, without doing too much work. Some of these errors my fit well into "events", but some will not.
>> 
>> ----- Original Message -----
>> From: "Maor Lipchuk" <mlipchuk at redhat.com>
>> To: "Yoav Kleinberger" <ykleinbe at redhat.com>
>> Cc: "Dan Kenigsberg" <danken at redhat.com>, devel at ovirt.org
>> Sent: Tuesday, May 13, 2014 5:26:32 PM
>> Subject: Re: [ovirt-devel] improving VDSM error reporting
>> 
>> The audit logs are presented in the events tab in the GUI, usually we
>> print an audit log when we succeed/fail to execute a command and usually
>> it is  more verbal then the CDA messages (Which presented in the dialog
>> boxes).
>> 
>> Since ConnectStorageDomain is a VDS command which runs in the execute
>> phase maybe it is more fitting to add an event log describing the error
>> in more details.
>> 
>> Regards,
>> Maor
>> 
>> On 05/13/2014 05:20 PM, Yoav Kleinberger wrote:
>>> Maor can you please explain? I'm not familiar with this audit log
>>> 
>>> ----- Original Message -----
>>> From: "Maor Lipchuk" <mlipchuk at redhat.com>
>>> To: "Yoav Kleinberger" <ykleinbe at redhat.com>, "Dan Kenigsberg" <danken at redhat.com>
>>> Cc: devel at ovirt.org
>>> Sent: Tuesday, May 13, 2014 5:17:02 PM
>>> Subject: Re: [ovirt-devel] improving VDSM error reporting
>>> 
>>> I would like to suggest another solution,
>>> Why not let the audit log present the extra error description.
>>> VDSM can override the existing error message before returning it to the
>>> engine with the extra data.
>>> 
>>> that way you will not have to add  new attributes to the existing API,
>>> and the dialog messages will still be localized.
>>> 
>>> regards,
>>> Maor
>>> 
>>> On 05/13/2014 04:50 PM, Yoav Kleinberger wrote:
>>>> well, we can add the python text, and the UX team will decide under what circumstances to display it. At least for some users it will add information.
>>>> 
>>>> ----- Original Message -----
>>>> From: "Dan Kenigsberg" <danken at redhat.com>
>>>> To: "Yoav Kleinberger" <ykleinbe at redhat.com>
>>>> Cc: devel at ovirt.org
>>>> Sent: Tuesday, May 13, 2014 4:08:47 PM
>>>> Subject: Re: [ovirt-devel] improving VDSM error reporting
>>>> 
>>>> On Tue, May 13, 2014 at 08:31:16AM -0400, Yoav Kleinberger wrote:
>>>>> Hi Dan,
>>>>> A few points:
>>>>> 
>>>>> 1. I explained the problem using the existing "message" field. An old VDSM will have the "meaningless" messages, and when working with a new engine they will be displayed to the user.
>>>> 
>>>> As I have shown, "message" is not completely meaningless as it is
>>>> (though it certainly does not add much information beyond the error
>>>> code). I don't suspect that Engine would show the extra "details" to
>>>> just any user. UX folks like to keep the GUI clean and tidy, so only
>>>> experienced users with the right credentials would see them. In some
>>>> (most) cases, "message" would add no information beyond the error code
>>>> and its localized string. But as time passes, we can add more and more
>>>> "meat" to the message.
>>>> 
>>>>> 2. The patch you attached generates the message inside the code for a specific circumstance. However, I want to handle unexpected errors that fly out of the code in a generic way, without needing to understand each and every case. 
>>>> 
>>>> unexpected errors end up with code error 16 and message 'Unexpected
>>>> exception'. Personally, I'd love to replace this message with a
>>>> meaningful traceback.
>>>> 
>>>> Dan.
>>>> _______________________________________________
>>>> Devel mailing list
>>>> Devel at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/devel
>>>> 
>>> 
>> 
> 
> _______________________________________________
> Devel mailing list
> Devel at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/devel