[openstack-dev] [trove] Configuration API BP
Craig Vyvial
cp16net at gmail.com
Wed Nov 13 18:23:58 UTC 2013
I have updated the reviews related to configuration groups.
Trove - https://review.openstack.org/#/c/53168/
Python-TroveClient - https://review.openstack.org/#/c/53169/
Please review at your leisure.
TODOs:
* add pagination support for configuration on instances
* mark configuration groups as deleted instead of doing a hard delete in
the db.
Thanks,
Craig Vyvial
On Thu, Oct 3, 2013 at 3:48 PM, Craig Vyvial <cp16net at gmail.com> wrote:
> Oops forgot the link on BP for versioning templates.
>
>
> https://blueprints.launchpad.net/trove/+spec/configuration-templates-versionable
>
>
> On Thu, Oct 3, 2013 at 3:47 PM, Craig Vyvial <cp16net at gmail.com> wrote:
>
>> I have been trying to figure out where a call for the "default"
>> configuration should go. I just finished adding a method to get the
>> [mysqld] section via an api call but not sure where this should go yet.
>>
>> Currently i made it:
>> GET - /instance/{id}/configuration
>>
>> This kinda only half fits in the path here because it doesnt really
>> describe that this is the "default" configuration on the instance. On the
>> other hand, it shows that it is coupled to the instance because we need the
>> instance flavor to give what the current values are in the template applied
>> to the instance.
>>
>> Maybe other options could be:
>> GET - /instance/{id}/configuration/default
>> GET - /instance/{id}/defaultconfiguration
>> GET - /instance/{id}/default-configuration
>> GET - /configuration/default/instance/{id}
>>
>> Suggestions welcome on the path.
>>
>> There is some wonkiness showing this information to the user because of
>> the difference in the values used. [1] This example shows that the template
>> uses "50M" as a value applied and the configuration-group would apply the
>> value equivalent to 52428800. I dont think we should worry about this now
>> but this could lead to confusion by a user. If they are a power-user type
>> then they might understand compared to a entry level user.
>>
>> [1] https://gist.github.com/cp16net/6816691
>>
>>
>>
>> On Thu, Oct 3, 2013 at 2:36 PM, McReynolds, Auston <amcreynolds at ebay.com
>> > wrote:
>>
>>> If User X's existing instance is isolated from the change, but there's
>>> no snapshot/clone/versioning of the current settings on X's instance
>>> (via the trove database or jinja template), then how will
>>> GET /configurations/:id return the correct/current settings? Unless
>>> you're planning on communicating with the guest? There's nothing
>>> wrong with that approach, it's just not explicitly noted anywhere in
>>> the blueprint. For some reason I inferred that it would be handled
>>> like trove security-groups.
>>>
>> So this is a great point. There are talks about making the templating
>> versioned in some form or fashion. ekonetzk(irc) said he would write up a
>> BP around versioning.
>>
>>
>>>
>>> On a slightly different note: If the default template will not be
>>> represented as a default configuration-group from an api standpoint,
>>> then how will you support the ability for a user to enumerate the list
>>> of default configuration-group values for a service-type?
>>> GET /configurations/:id won't be applicable, so will it be
>>> something like GET /configurations/default?
>>>
>> see above paragraph.
>>
>>
>>>
>>>
>>>
>>> From: Craig Vyvial <cp16net at gmail.com>
>>> Reply-To: OpenStack Development Mailing List
>>> <openstack-dev at lists.openstack.org>
>>> Date: Thursday, October 3, 2013 11:17 AM
>>> To: OpenStack Development Mailing List <
>>> openstack-dev at lists.openstack.org>
>>> Subject: Re: [openstack-dev] [trove] Configuration API BP
>>>
>>>
>>> inline.
>>>
>>>
>>> On Wed, Oct 2, 2013 at 1:03 PM, McReynolds, Auston
>>> <amcreynolds at ebay.com> wrote:
>>>
>>> Awesome! I only have one follow-up question:
>>>
>>> Regarding #6 & #7, how will the clone behavior work given that the
>>> defaults are hydrated by a non-versioned jinja template?
>>>
>>>
>>> I am not sure i understand "clone behavior" because there is not really a
>>> concept of cloning here. The jinja template is created and passed in the
>>> "prepare call" to the guest to write to the default my.cnf file.
>>>
>>> When a configuration-group is removed the instance will return to the
>>> "default" state. This does not exactly act as a clone behavior.
>>>
>>>
>>>
>>> Scenario Timeline:
>>>
>>> T1) Cloud provider begins with the default jinja template, but changes
>>> the values for properties 'a' and 'b'. (Template Version #1)
>>> T2) User X deploys a database instance
>>> T3) Cloud provider decides to update the existing template by modifying
>>> property 'c'. (Template Version #2)
>>> T4) User Z deploys a database instance
>>>
>>> I think it goes without saying that User Z's instance gets Template
>>> Version #2 (w/ changes to a & b & c), but does User X?
>>>
>>>
>>> No User X does not get the changes. For User X to get the changes a
>>> maintenance may need to be scheduled.
>>>
>>>
>>>
>>> If it's a "true" clone, User X should be isolated from a change in
>>> defaults, no?
>>>
>>>
>>> User X will not see these default changes until a new instance is
>>> created.
>>>
>>>
>>>
>>> Come to think about it, this is eerily similar to security-groups:
>>> administratively, it can be beneficial to share a
>>> configuration/security-group across multiple instances, but it can
>>> also be a nightmare. Internally, it's extremely rare that we wish to
>>> apply a database change to multiple tenants at once, so I'd argue
>>> at a minimum to support a CONF opt-in for isolation, if not default
>>> to it.
>>>
>>>
>>> If i understand this correctly my above statement means that its isolated
>>> by default.
>>>
>>>
>>>
>>> On a related note: Will the default template for a service-type be
>>> represented as a default configuration-group? If so, I imagine it
>>> can be managed through the API (or MGMT API)?
>>>
>>>
>>> The default template will not be represented as a configuration group.
>>> This could potentially be a good fit but its more of a nice to have type
>>> of feature.
>>>
>>>
>>>
>>>
>>> From: Craig Vyvial <cp16net at gmail.com>
>>> Reply-To: OpenStack Development Mailing List
>>> <openstack-dev at lists.openstack.org>
>>>
>>> Date: Wednesday, October 2, 2013 10:06 AM
>>> To: OpenStack Development Mailing List <
>>> openstack-dev at lists.openstack.org>
>>>
>>> Subject: Re: [openstack-dev] [trove] Configuration API BP
>>>
>>>
>>> I'm glad we both agree on most of these answers.
>>> :)
>>>
>>> On Oct 2, 2013 11:57 AM, "Michael Basnight" <mbasnight at gmail.com> wrote:
>>>
>>> On Oct 1, 2013, at 11:20 PM, McReynolds, Auston wrote:
>>>
>>> > I have a few questions left unanswered by the blueprint/wiki:
>>> >
>>> > #1 - Should the true default configuration-group for a service-type be
>>> > customizable by the cloud provider?
>>>
>>> Yes
>>>
>>> >
>>> > #2 - Should a user be able to enumerate the entire actualized/realized
>>> > set of values for a configuration-group, or just the overrides?
>>>
>>> actualized
>>>
>>> >
>>> > #3 - Should a user be able to apply a different configuration-group on
>>> > a master, than say, a slave?
>>>
>>> Yes
>>>
>>> >
>>> > #4 - If a user creates a new configuration-group with values equal to
>>> > that of the default configuration-group, what is the expected
>>> > behavior?
>>>
>>> Im not sure thats an issue. You will select your config group, and it
>>> will
>>> be the one used. I believe you are talking the difference between the
>>> "template" thats used to set up values for the instance, and the config
>>> options that users are allowed to edit.
>>> Those are going to be "appended", so to speak, to the existing template.
>>> Itll be up to the server software to define what order values, if
>>> duplicated, are read / used.
>>>
>>> >
>>> > #5 - For GET /configuration/parameters, where is the list of supported
>>> > parameters and their metadata sourced from?
>>>
>>>
>>>
>>> i believe its a db tableŠ someone may have to correct me there.
>>>
>>> >
>>> > #6 - Should a user be able to reset a configuration-group to the
>>> > current default configuration-group?
>>>
>>> Yes, assuming we have a "default config group", and im not sure we have a
>>> concept of that. We have what the install creates, the templated config
>>> file. Removing the association of your config from the instance will do
>>> this thought.
>>>
>>> >
>>> > #7 - Is a new configuration-group a clone of the then current default
>>> > configuration-group with various changes, or will inheritence be
>>> > utilized?
>>>
>>> I think clone will be saner for now. But you can edit your group with a
>>> PATCH, and that will not clone it. See [1] first paragraph.
>>>
>>> >
>>> > #8 - How should the state of pending configuration-group changes be
>>> > reflected in GET /instances/:id ? How will this state be
>>> > persisted?
>>>
>>> You are talking about changes that require a restart i believe. I think
>>> this falls into the same category as our conversation about minor version
>>> updates. We can have a pretty generic "restart required" somewhere there.
>>>
>>> >
>>> > #9 - Reminder: Once multiple service-types and versions are supported,
>>> > the configuration-group will need a service-type field.
>>>
>>> Most def. You will only be able to assign relevant configs to their
>>> service-types, and the /configuration/parameters will need to be typed
>>> too.
>>>
>>> >
>>> > #10 - Should dynamic values (via functions and operators) in
>>> > configuration-groups be supported?
>>> > Example: innodb_buffer_pool_size = 150 * flavor['ram']/512
>>>
>>> Hmmmm. This is quite interesting. But no, not v1. I totally agree w/ the
>>> nice-to-have. Good idea though, we should add it to the blueprint.
>>>
>>> >
>>> > My Thoughts:
>>> >
>>> > #1 - Yes
>>> > #2 - Actualized
>>> > #3 - Yes
>>> > #4 - Depends on whether the approach for configuration-groups is to
>>> > clone or to inherit.
>>> > #5 - ?
>>> > #6 - Yes
>>> > #7 - ?
>>> > #8 - ?
>>> > #9 - N/A
>>> > #10 - In the first iteration of this feature I don't think it's an
>>> > absolute necessity, but it's definitely a nice-to-have. The
>>> > design/implementation should not preclude this from being
>>> easily
>>> > added in the future.
>>> >
>>> > Where "?" == "I'd like to think about it a bit more, but I have a gut
>>> > feeling"
>>> >
>>> > Thoughts?
>>>
>>> [1]
>>>
>>> http://lists.openstack.org/pipermail/openstack-dev/2013-October/015919.html
>>>
>>>
>>> <
>>> http://lists.openstack.org/pipermail/openstack-dev/2013-October/015919.ht
>>> ml<http://lists.openstack.org/pipermail/openstack-dev/2013-October/015919.html>
>>> <
>>> http://lists.openstack.org/pipermail/openstack-dev/2013-October/015919.htm
>>> l>>
>>>
>>> _______________________________________________
>>> OpenStack-dev mailing list
>>> OpenStack-dev at lists.openstack.org
>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>
>>>
>>>
>>> _______________________________________________
>>> OpenStack-dev mailing list
>>> OpenStack-dev at lists.openstack.org
>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> OpenStack-dev mailing list
>>> OpenStack-dev at lists.openstack.org
>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20131113/0db3abfc/attachment.html>
More information about the OpenStack-dev
mailing list