fedmsg consumer for awarding open badges
This repo contains the consumer and the command necessary to hook the badges stack (Tahrir, Tahrir-API, Tahrir-REST) into fedmsg. It is the process that runs in the background, monitoring activity of Fedora contributors, and is responsible for awarding badges for activity as it happens. It is separate from and sometimes confused with the frontend of the badges system; that web application is called tahrir. This project (fedbadges) writes to a database that the web frontend (tahrir) reads from.
The actual badge rules that we act on in Fedora Infrastructure can be found here in our ansible repository.
fedbadges runs as a Consumer plugin to the fedmsg-hub (really, a moksha-hub). When started, it will load some initial configuration and a set of BadgeRules (more on that later) and then sit quietly listening to the fedmsg bus. Each rule (composed of some metadata, a trigger, and a set of criteria) is defined on disk as a yaml file.
When a new message comes along, our Consumer looks to see if it matches any of the BadgeRules it has registered.
Each BadgeRule must define a trigger – a
lightweightcheck. When processing a message, this is the first thing that is checked. It defines a pattern that the message must match. If the message does not match, then the current BadgeRule is discarded and processing moves to the next.
A trigger is typically something like “any bodhi message” or “messages only from the failure of a koji build”. More on their specification below.
BadgeRules must also define a set of criteria – a more
heavyweightchecks. During the processing of a newly received message, if the message matches a BadgeRule’s trigger, the criteria is then considered. This typically involves a more expensive query to the datanommer database.
A BadgeRule criteria may read something like “$user has pushed 200 bodhi updates to stable” or “$user chaired an IRC meeting”.
Aside: Although datanommer is the only currently supported backend, we can implement other queryable backend in the future as needed like FAS2 (to see if the user is in X number of groups) or even off-site services like libravatar (to award a badge if the user is a user of the AGPL web service).
If a badge’s trigger and criteria both match, then the badge is awarded. If the BadgeRule doesn’t specify, we award the badge to all usernames returned by a call to fedmsg.meta.msg2usernames(msg).
That is usually correct – but sometimes, a BadgeRule needs to specify that one particular user (not all related users) should be recipients of the badge. In this case, the BadgeRule may define a recipient in dot-notation that instructs the Consumer how to extract the recipient’s username from the received message.
There are some optimizations in place omitted above for clarity. For instance, after the trigger has matched we first check if the user that would be awarded the badge already has it. If they do, we stop processing the badge rule immediately to avoid making an unnecessary expensive check against the datanommer db.
Configuration - Global
fedbadges needs three major pieces of global configuration. All configuration is loaded in the standard fedmsg way, from python files in /etc/fedmsg.d/.
First, generic and tahrir-related configuration. See fedmsg.d/badges-global.py in the git repo for an example.
Second, datanommer connection information. See fedmsg.d/datanommer.py in the git repo for an example.
Third, fedbadges emits its own fedmsg messages when it awards badges. It will need its own endpoint definitions for this. See fedmsg.d/endpoints.py in the git repo for an example.
Configuration - BadgeRule specification
BadgeRules are specified in YAML on the file system.
Every BadgeRule must carry the following minimum set of metadata:
# This is some metadata about the badge name: Like a Rock description: You have pushed 500 or more bodhi updates to stable status. creator: ralph # This is a link to the discussion about adopting this as a for-real badge. discussion: http://github.com/fedora-infra/badges/pull/SOME_NUMBER # A link to the image for the badge image_url: http://somelink.org/to-an-image.png
Here’s a simple example of a trigger:
trigger: category: bodhi
The above will match any bodhi message on any of the topics that come from the bodhi update system.
Triggers may employ a little bit of logic to make more complex filters. The following trigger will match any message that comes from either the bodhi update system or the fedora git package repos:
trigger: category: any: - bodhi - git
At present triggers may directly compare themselves against only the
category or the
topic of a message. In the future we’d like to add
more comparisons.. in the meantime, here’s an example of comparing against
the fully qualified message topic. This will match any message
that is specifically for editing a wiki page:
trigger: topic: org.fedoraproject.prod.wiki.article.edit
There is one additional way you can specify a trigger. If you need more flexibility than topic and category allow, you may specify a custom filter expression with a lambda filter. For example:
trigger: lambda: "a string of interest" in json.dumps(msg)
The above trigger will match if the string "a string of interest" appears anywhere in the incoming message. fedbadges takes the expression you provide it and compiles it into a python callable on initialization. Our callable here serializes the message to a JSON string before doing its comparison. Powerful!
As mentioned above in the architecture section, we currently only support datanommer as a queryable backend for criteria. We hope to expand that in the future.
Datanommer criteria are composed of three things:
- A filter limits the scope of the query to datanommer.
- An operation defines what we want to do with the filtered query. Currently, we can only count the results.
- A condition defines how we want to compare the results of the operation to determine if our criteria matches or not.
Here’s an example of a simple criteria definition:
criteria: filter: topics: - "%(topic)s" operation: count condition: greater than or equal to: 2
The above criteria will match if there is more than one message in datanommer
with the same topic as the incoming message being handled. Here, "%(topic)s"
template variable. Template variables will have their values
substituted before the expensive check is made against datanommer.
The above example doesn’t make much sense – we’d never use it for a real badge. The criteria would be true if there were two of any message kicked off by any user at any time in the past. Pretty generic. Here’s a more interesting criteria definition:
criteria: filter: topics: - org.fedoraproject.prod.git.receive usernames: - "%(msg.commit.username)s" operation: count condition: greater than or equal to: 50
This criteria would match if there existed 50 messages of the topic "org.fedoraproject.prod.git.receive" that were also kicked off by whatever user is listed in the msg['msg']['commit']['username'] field of the message being currently processed. In other words, this criteria would match if the user has pushed to the fedora git repos 50 or more times.
You can do some fancy things with the condition of a datanommer filter. Here’s a list of the possible comparisons you can make:
- "is greater than or equal to" or alternatively "greater than or equal to"
- "greater than"
- "is less than or equal to" or alternatively "less than or equal to"
- "less than"
- "equal to" or alternatively "is equal to"
- "is not" or alternatively "is not equal to"
As you can see, some of them are synonyms for each other.
If any of those don’t meet your needs, you can specify a custom expression by using the lambda condition whereby fedbadges will compile whatever statement you provide into a callable and use that at runtime. For example:
criteria: filter: topics: - org.fedoraproject.prod.git.receive usernames: - "%(msg.commit.username)s" operation: count condition: lambda: value != 0 and ((value & (value - 1)) == 0)
Who knows why you would want to do this, but the above criteria check will succeed if the number of messages returned from the filtered datanommer query is exactly a power of 2.
By default, if the trigger and criteria match, fedbadges will award badges to all the users returned by a call to fedmsg.meta.msg2usernames(msg). This usually corresponds with “what users are responsible” for this message. That is usually what we want to award badges for.
There are some instances for which that is not what we want.
Take the org.fedoraproject.prod.fas.group.member.remove message for example. When user A removes user B from a group, both usernames are returned from a call to fedmsg.meta.msg2usernames(msg) with no further distinction as to which was removing and which was removed.
Imagine we have a “Group Pruner” badge that’s awarded to group admins who remove inactive users from groups. We don’t want to inadvertently award that badge to the persons who were removed, only to those who removed them.
To allow for this scenario, badges may optionally define a recipient in dotted notation that tells fedbadges where to find the username of the recipient in the originating message. For instance, the following would handle the fas case we described above:
trigger: topic: org.fedoraproject.prod.fas.group.member.remove criteria: filter: topics: - "%(topic)s" operation: count condition: greater than or equal to: 1 recipient: "%(msg.agent.username)s"