Conversation
| def trigger_hooks(event, details) | ||
| BarclampCatalog.barclamps.keys.each do |barclamp| | ||
| begin | ||
| cls = eval("#{barclamp.camelize}Service") |
There was a problem hiding this comment.
Lint/Eval: The use of eval is a serious security risk.
matelakat
left a comment
There was a problem hiding this comment.
Thanks for the clarification! Please see my comment about the @notify_on_save hash.
| end | ||
| # deep clone of @role.default_attributes, used when saving node | ||
| @attrs_last_saved = @role.default_attributes.deep_dup | ||
| @notify_on_save = {} |
There was a problem hiding this comment.
This variable really holds the notifications to be sent on the next save, if understand it correctly. Also looking at the point where the notifications are sent out, it seems, that we are not using the value in the hash, only the keys. I think a better name for this variable might help, and also making it an array/set would also help. I would name it something like: on_save_events. Thus the protocol would look like: the client adds/queues up items to the on_save_events collection, and those are being sent and flushed on save.
| def update_public_name(value) | ||
| unless value.nil? | ||
| crowbar["crowbar"]["public_name"] = value | ||
| @notify_on_save["public_name"] = true |
There was a problem hiding this comment.
It might be a stupid question: Are we sure that update_public_name is only called whenever the value of it has changed? I'm not seeing the same check-if-changed mechanism as earlier.
There was a problem hiding this comment.
See https://github.com/crowbar/crowbar-core/pull/952/files#diff-496ddf4ffc5694a5fcdddbff8206dc8dR441 (there's also force_public_name= where the check is missing, but I think it's fine to still send the event in that case since it's forced).
|
|
||
| details = { | ||
| node: @node.name, | ||
| attributes: @notify_on_save.keys |
There was a problem hiding this comment.
As I mentioned earlier, we only send the keys, not the values. So we might want to send the values as well, or use a single set as mentioned earlier.
There was a problem hiding this comment.
I can move to a set, sure. FWIW, in ruby, a set is implemented (behind the scenes) exactly the way I did, so won't be too much different but I guess it's slightly more readable.
66792a7 to
2ee29ec
Compare
| def trigger_hooks(event, details) | ||
| BarclampCatalog.barclamps.keys.each do |barclamp| | ||
| begin | ||
| cls = eval("#{barclamp.camelize}Service") |
There was a problem hiding this comment.
Lint/Eval: The use of eval is a serious security risk.
There was a problem hiding this comment.
Calculating the class name and obtaining the class surely belongs in methods in BarclampCatalog. Doing it here sounds like privacy violation / feature envy.
There was a problem hiding this comment.
Isnt the way of doing this better with Kernel.const_get ?*
*not a ruby expert
There was a problem hiding this comment.
We actually have ServiceObject.get_service exactly for this...
|
@matelakat I addressed your comments. I also renamed the small library to EventDispatcher as Notify was too generic... |
| end | ||
| end | ||
| end | ||
| handle_asynchronously :trigger_hooks |
There was a problem hiding this comment.
Should be a blank line before this.
There was a problem hiding this comment.
Oh, maybe not. It's kind of like a decorator.
| proposals.each do |proposal| | ||
| role = proposal.role | ||
| next if role.nil? | ||
| service.event_hook(role, event, details) |
There was a problem hiding this comment.
We need some exception catching here.
| @node["crowbar_wall"] ||= {} | ||
| @node["crowbar_wall"]["openstack"] ||= {} | ||
| @node["crowbar_wall"]["openstack"]["availability_zone"] = value | ||
| if @node["crowbar_wall"]["openstack"]["availability_zone"] != value |
There was a problem hiding this comment.
@vuntz as I mentioned it in the "eliminate saves" PR earlier, this pattern might worth extracting to a helper, something like: set_value_and_call_function_if_changed or I believe ruby must have better ways to express it.
| } | ||
| Crowbar::EventDispatcher.trigger_hooks(:node_changed, details) | ||
|
|
||
| @on_save_events = Set.new |
There was a problem hiding this comment.
Can this be called in parallel by different puma threads? If so isn't this a race and we could miss save events when cleaning the Set?
There was a problem hiding this comment.
The node object is not shared among puma threads, so it should be fine. When we create threads ourselves, we should be careful of course and use proper locking mechanisms, but it's not specific to this change.
1fee9e4 to
83ad9a3
Compare
aspiers
left a comment
There was a problem hiding this comment.
Once the variable is renamed and comment addressed, I'm good with this PR.
| end | ||
| # deep clone of @role.default_attributes, used when saving node | ||
| @attrs_last_saved = @role.default_attributes.deep_dup | ||
| @attributes_changed_for_event = Set.new |
There was a problem hiding this comment.
As discussed could you rename this to @attributes_changed and then add a comment explaining what it's used for?
|
Well, of course the merge conflict and Travis failures have to be fixed first :) |
There are several cases where we have events that should trigger some
activity in some other part of Crowbar. For instance:
- when the public name of a node is saved, it may impact the endpoint
of an OpenStack service
- when the public name of the VIP of haproxy is changed, it impacts the
endpoint of OpenStack services
- when the keystone proposal is applied, we may want to reapply all
proposals that depend on keystone
What we need for this is the ability to notify about the events in the
rails application and then to dispatch the notifications to hooks listen
listening to them that will decide if some action should be triggered.
The main reason we didn't have this in the past is that we likely don't
want to do that in the foreground of the rails application. But now that
we have delayed_job, we can send the notifications and run the hooks in
the background.
In this commit, we add the simple infrastructure about notifications and
hooks:
- the events are defined with a name and a hash that contains the
details of the event. The structure of the hash depends on the event.
- we have two events that are generated and notified: node_changed
(when a node is saved and some specific attributes have been changed)
and proposal_applied (when a proposal is successfully applied).
- a simple dispatcher exists that simply connects the hooks to the
events.
- the hooks only exist for service objects for the time being; a
service object simply needs to have a event_hook method to register
the hook, and will need to filter for the events it cares about. The
signature of event_hook is as follows:
def event_hook(role, event, details)
It could be argued that the hooks should be registered for some specific
events (hence moving the filter to the event dispatcher), but it's not
worth the complexity for now.
We could add many more events but for the time being, we only add the
events that we know are useful.
|
@aspiers rebased |
Rebase of PR crowbar#952 Only committing the infrastructure changes For usage example(s), see: crowbar#952 crowbar/crowbar-ha#171 crowbar/crowbar-openstack#717 There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance: - when the public name of a node is saved, it may impact the endpoint of an OpenStack service - when the public name of the VIP of haproxy is changed, it impacts the endpoint of OpenStack services - when the keystone proposal is applied, we may want to reapply all proposals that depend on keystone What we need for this is the ability to notify about the events in the rails application and then dispatch the notifications to hooks which listen for them and decide if some action should be triggered. The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background. In this commit, we add the simple infrastructure about notifications and hooks: - the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event. - a simple dispatcher exists that simply connects the hooks to the events. - the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details) It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
Rebase of PR crowbar#952 but only including the new infrastructure class For usage examples, see: crowbar#952 crowbar/crowbar-ha#171 crowbar/crowbar-openstack#717 There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance: - when the public name of a node is saved, it may impact the endpoint of an OpenStack service - when the public name of the VIP of haproxy is changed, it impacts the endpoint of OpenStack services - when the keystone proposal is applied, we may want to reapply all proposals that depend on keystone What we need for this is the ability to notify about the events in the rails application and then dispatch the notifications to hooks which listen for them and decide if some action should be triggered. The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background. In this commit, we add the simple infrastructure about notifications and hooks: - the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event. - a simple dispatcher exists that simply connects the hooks to the events. - the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details) It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
Rebase of PR crowbar#952 but only including the new infrastructure class For usage examples, see: crowbar#952 crowbar/crowbar-ha#171 crowbar/crowbar-openstack#717 There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance: - when the public name of a node is saved, it may impact the endpoint of an OpenStack service - when the public name of the VIP of haproxy is changed, it impacts the endpoint of OpenStack services - when the keystone proposal is applied, we may want to reapply all proposals that depend on keystone What we need for this is the ability to notify about the events in the rails application and then dispatch the notifications to hooks which listen for them and decide if some action should be triggered. The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background. In this commit, we add the simple infrastructure about notifications and hooks: - the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event. - a simple dispatcher exists that simply connects the hooks to the events. - the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details) It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
Rebase of PR crowbar#952 but only including the new infrastructure class For usage examples, see: crowbar#952 crowbar/crowbar-ha#171 crowbar/crowbar-openstack#717 There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance: - when the public name of a node is saved, it may impact the endpoint of an OpenStack service - when the public name of the VIP of haproxy is changed, it impacts the endpoint of OpenStack services - when the keystone proposal is applied, we may want to reapply all proposals that depend on keystone What we need for this is the ability to notify about the events in the rails application and then dispatch the notifications to hooks which listen for them and decide if some action should be triggered. The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background. In this commit, we add the simple infrastructure about notifications and hooks: - the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event. - a simple dispatcher exists that simply connects the hooks to the events. - the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details) It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
Rebase of PR crowbar#952 but only including the new infrastructure class For usage examples, see: crowbar#952 crowbar/crowbar-ha#171 crowbar/crowbar-openstack#717 There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance: - when the public name of a node is saved, it may impact the endpoint of an OpenStack service - when the public name of the VIP of haproxy is changed, it impacts the endpoint of OpenStack services - when the keystone proposal is applied, we may want to reapply all proposals that depend on keystone What we need for this is the ability to notify about the events in the rails application and then dispatch the notifications to hooks which listen for them and decide if some action should be triggered. The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background. In this commit, we add the simple infrastructure about notifications and hooks: - the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event. - a simple dispatcher exists that simply connects the hooks to the events. - the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details) It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
Rebase of PR crowbar#952 but only including the new infrastructure class For usage examples, see: crowbar#952 crowbar/crowbar-ha#171 crowbar/crowbar-openstack#717 There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance: - when the public name of a node is saved, it may impact the endpoint of an OpenStack service - when the public name of the VIP of haproxy is changed, it impacts the endpoint of OpenStack services - when the keystone proposal is applied, we may want to reapply all proposals that depend on keystone What we need for this is the ability to notify about the events in the rails application and then dispatch the notifications to hooks which listen for them and decide if some action should be triggered. The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background. In this commit, we add the simple infrastructure about notifications and hooks: - the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event. - a simple dispatcher exists that simply connects the hooks to the events. - the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details) It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
|
Completed in #1308 |
There are several cases where we have events that should trigger some activity in some other part of Crowbar. For instance:
What we need for this is the ability to notify about the events in the rails application and then to dispatch the notifications to hooks listen listening to them that will decide if some action should be triggered.
The main reason we didn't have this in the past is that we likely don't want to do that in the foreground of the rails application. But now that we have delayed_job, we can send the notifications and run the hooks in the background.
In this commit, we add the simple infrastructure about notifications and hooks:
the events are defined with a name and a hash that contains the details of the event. The structure of the hash depends on the event.
we have two events that are generated and notified: node_changed (when a node is saved and some specific attributes have been changed) and proposal_applied (when a proposal is successfully applied).
a simple dispatcher exists that simply connects the hooks to the events.
the hooks only exist for service objects for the time being; a service object simply needs to have a event_hook method to register the hook, and will need to filter for the events it cares about. The signature of event_hook is as follows: def event_hook(role, event, details)
It could be argued that the hooks should be registered for some specific events (hence moving the filter to the event dispatcher), but it's not worth the complexity for now.
We could add many more events but for the time being, we only add the events that we know are useful.