On Wed, Oct 10, 2018 at 1:52 PM Marcin Sobczyk <msobczyk@redhat.com> wrote:


On 10/10/18 10:14 AM, Nir Soffer wrote:
Adding devel.

---------- Forwarded message ---------
From: Nir Soffer <nsoffer@redhat.com>
Date: Wed, 10 Oct 2018, 11:07
Subject: Re: VDSM API schema inconsistencies
To: Piotr Kliczewski <pkliczew@redhat.com>
Cc: Dan Kenigsberg <danken@redhat.com>, Edward Haas <ehaas@redhat.com>, Milan Zamazal <mzamazal@redhat.com>, Tomasz Baranski <tbaransk@redhat.com>, Francesco Romani <fromani@redhat.com>, Irit Goihman <igoihman@redhat.com>, Marcin Sobczyk <msobczyk@redhat.com>




On Wed, 10 Oct 2018, 10:21 Piotr Kliczewski, <pkliczew@redhat.com> wrote:
+Nir

On Wed, Oct 10, 2018 at 8:39 AM Marcin Sobczyk <msobczyk@redhat.com> wrote:
Hi,
I would love to get some feedback from you about it on the VDSM call today so if you have a moment, please take a look.
Marcin

---------- Forwarded message ---------
From: Marcin Sobczyk <msobczyk@redhat.com>
Date: Mon, Oct 8, 2018 at 2:40 PM
Subject: VDSM API schema inconsistencies
To: devel <devel@ovirt.org>


Hi,

when I was working on BZ#1624306 it turned out, that even if I'd fix the help-printing code in vdsm-client, there's still a lot of inconsistencies in API's schema files.

After some investigation, I found that there's a code for reporting schema inconsistencies, but it's turned off because it bloats the logs.

Right, an it should remain like this. In runtime
we never want these warnings.

There's also a "strict" mode that raises an exception each time an inconsistency is found, but it's also off because there's so many of them.


Same.

Both are debugging aids for developers, to
make it easy to check the schema.

I think that no one wants to maintain both: the actual code and schema files.


This is not a big issue. We rarely add or 
change APIs. But it is nice if we don't need to
manualy make the schema correct.

My idea would be to make the schema derived from the code at build time as a long-term effort.


This cannot work, since clients depend on the schema, it should not change if when someone
makes uintended change in the code.

It should work in the opposite way, you change the schema, and generate new code frome schema.

Threre are existing tools like grpc and thrift that
work in this way. For long term we should drop
our jsonrpc/stomp stack and move to grpc.
Switching to grpc or thrift is a great idea, but that's also a great effort.

For short term we can just fix the schema to
reflect actual engine interaction. when engine
and schema do not match, engine is usualy
right. For responses, if engine can handle the
response the code is right.

Please do not make RPC code more complex
and inefficient for improving schema issues
detection. I would rather remove this code
instead.

 My goal is quite opposite:

  • remove as much code as possible
Yes!
 
  • switch to something more standard like JSON Schema for validation
Yes!

But I think json schema is quite slow, so we need to be careful if we use it
in runtime.
 
  • do the validation only in our test suites and not the client code
In the test you can check that the code matches the schema (e.g. known
parameter is accepted, missing required argument cause a failure), but
calling most APIs is hard and have unwanted side effects, so this can be
tested only in the api layer, checking that the code call the next layer.

Testing the response is mostly impossible in tests, since it depends on
the actual state of the system which is hard to setup in the tests.
It would be useful if we can add validation with the schema for existing
tests that check responses. I don't think we have many tests like this.
 
  • catch errors ASAP - preferably at build time, without having to run any code at all
Build time should run any tests, for this we have automated tests.
Maybe my intention was not clear enough, so I've assembled a simple PoC: http://pastebin.test.redhat.com/655651
To run it you'll need to install 'typed-ast' in your venv.

Before we can do that though, we need to address the inconsistencies first.

https://gerrit.ovirt.org/#/q/status:open+project:vdsm+branch:master+topic:vdsm-api-schema-debug

This is a series of patches that helps fixing schema issues. What it does, is it adds a bunch of information from stack to vastly ease the process of fixing them. The logging level is also changed from 'WARNING' to 'DEBUG'. Here's an example of a logged entry:

Host.setKsmTune
With message: Required property pages_to_scan is not provided when calling Host.setKsmTune
With backtrace: [
        [
                "_verify_object_type",
                {
                        "call_arg_keys":[
                                "merge_across_nodes",
                                "run"
                        ],
                        "schema_type_name":"KsmTuneParams"
                }
        ],
        [
                "_verify_complex_type",
                {
                        "schema_type_type":"object"
                }
        ]
]

To make it work, a patch for OST's 'basic-master-suite' is on the way that switches 'schema.inconsistency' logger's logging level do 'DEBUG'. That way, we can find all reported errors in 'vdsm-schema-error.log' files.

An initial report that groups reported errors by namespaces/methods has also been made. Please note, that an implementation that completely avoids error duplicates is really hard and IMHO not worth the effort. You can find the results here:


File a bug per vertical (infra, storage, virt) for
these issues. Since we don't know about real user issues with these apis this must be incorrect schema.
Will do so.

Nir

_______________________________________________
Devel mailing list -- devel@ovirt.org
To unsubscribe send an email to devel-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/site/privacy-policy/
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/devel@ovirt.org/message/JHTGTSN376HHJHNCNRMGIIXAQMFT2YQ6/