Stop randomize_child_id when handoff #226
Conversation
randomize_child_id adds a new (duplicated) process to DynamicSupervisor.DeltaCrdt which results to a keep growing list of processes
kunna
changed the title
|
this makes sense to solve #207 can i know why was this closed? |
|
I agree. without this change it is effectively not possible to run a Horde.DynamicSupervisor without a Horde.Registry... |
|
I am not sure if it is possible to run a Horde.DynamicSupervisor without a Horde.Registry even with this change. |
|
Not sure if I follow.. why would that be? Assuming that this change fixes the process duplications on membership changes.. |
|
@arjan because there are other situations in which you can end up with duplicate processes. A netsplit, for example. |
|
In a previous version, Horde.DynamicSupervisor was deduping processes, but this resulted in race conditions with Horde.Registry, resulting in both processes being removed sometimes. That's why Horde.DynamicSupervisor doesn't dedupe processes anymore. |
resolves #207
Debug
Following the instruction from #207, I noticed that new processes are keep added to the
DeltaCrdt.read(Conv.ConversationSupervisor.Crdt). Whenever a new process is handed off to another node, a duplicate record is created, with a new child.idChange
By removing the
randomize_child_idduring the handoff process, I was able to fix this issue.Question
I try to understand why we needed to add
randomize_child_idand found this PR. #196This PR says, "Deduplication will be the sole responsibility of Horde.Registry", but I was not able to find the logic from the
Horde.Registryfixing the duplicated processes.I think PR #196 was meant to remove
Horde.ProcessesSupervisor.terminate_child_by_id. https://github.com/derekkraan/horde/pull/196/files#diff-36c0ba7f42365d4eeaa857f4b899986cb28d29ff94de730fd8c1934c872a0389L476because this change stoped terminating the old process, we need a consistent id for
Horde.Registryto prevent creating a duplicated process.