JarCastingakka-kadmin
The kadmin actor will perform a best effort deduplication. To do so the code assumes there is only one kadmin actor interacting with kerberos.
By best effort we mean that there is still a very small window in which duplication can still occur. To understand it, we must first explain how the code operates: The kadmin actor knows the last deliveryId it handled for a given sender. When a new message arrives it will check its deliveryId against the expectedId (last deliveryId + 1) of that sender.
-
If (deliveryId > expectedId) then
The message is a future one and there are unprocessed messages in between, so the actor will ignore it.
-
If (deliveryId < expectedId) then
The message is a resend, the actor will just respond with its result. This result was stored in the actor internal state when the message side-effect was performed.
-
If (deliveryId == expectedId) then
This is the message we are expecting, so the side-effect will performed the following way:
- Store in memory that we started performing the side-effect.
- Perform the side-effect.
- Persist the side-effect result, and then in the persist handler, update the actor internal state with the result.
If the actor crashes:
- After performing the side-effect
- But before persisting its result
Duplication will occur because when recovering the actor won't know that it already performed the side-effect. This is also true if we persisted that we started executing the side-effect.
For most operations this won't be a problem because they are idempotent. However for ChangePassword and consequently AddPrincipal (since ChangePassword might be invoked inside AddPrincipal) this might be a problem because these operations might not be idempotent in some cases.
Install
Add the following dependency to your build.sbt:
libraryDependencies += "pt.tecnico.dsi" %% "akka-kadmin" % "0.5.0"
We use semantic versioning.
Configurations
akka-kadmin uses typesafe-config.
The reference.conf file has the following keys:
akka-kadmin {
# How long to wait to actually perform the remove.
# It is useful to perform the remove at a later time to deal with delayed messages.
# If set to 0 then the remove will be performed instantaneously
remove-delay = 15 minutes
# Every X messages a snapshot will be saved. Set to 0 to disable automatic saving of snapshots.
# This value is not a strict one because the kadmin actor will not persist it.
# If 190 messages were already processed and then the actor crashs only after
# another 200 messages will the actor save a snapshot.
# Repeated and future messages will not count. Or, in other words, only messages where the deliveryId
# is equal to the expectedId will increase the counter.
save-snapshot-every-X-messages = 200
# Akka-kadmin will use as settings for kadmin library those defined:
# 1) Here, directly under the path akka-kadmin (these have precedence over the next ones).
# 2) On the same level as akka-kadmin.
}
Alternatively you can pass your Config object to the Kadmin actor directly, or subclass the Settings class for a mixed approach. The scaladoc of the Settings class has examples explaining the different options.
How to test akka-kadmin
In the project root run ./test.sh. This script will run docker-compose up inside the docker-kerberos folder. Be sure to have docker and docker-compose installed on your computer.
Note on the docker-kerberos folder
This folder is a git fake submodule to the docker-kerberos repository.
License
akka-kadmin is open source and available under the MIT license.
