[
https://issues.apache.org/jira/browse/MESOS-1420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14011905#comment-14011905
]
Adam B commented on MESOS-1420:
-------------------------------
While I agree that current ids (and paths) are lengthy, they are built on top
of the MasterID, which is of the semi-human-readable format:
startdate/time-ip-port-upid. FrameworkIDs, SlaveIDs, and OfferIDs are just the
MasterID which created/registered that Fwk/Slave/Offer, appended with a counter
value. I would propose that we prepend an 'f', 's', or 'o' to the counter
value, so that the Nth framework, slave, and offer can actually be
distinguished from each other by their IDs.
Given that Fwk/Slave/OfferIDs are based on a MasterID, we can probably shorten
the MasterID and still keep it unique and somewhat human-readable. For example,
you shouldn't need both ip-port and upid, only one is necessary. And maybe the
date could just be in seconds since 1970 (`date +%s`), which results in 11
chars instead of 19.
The "runs" ID is actually just the containerID, created as UUID::random()
during Framework::launchExecutor, but we would prefer that the containerizer
itself be able to set the containerID, which means that it could be anything
(Docker container IDs?).
> Shorten slave, framework and run IDs
> ------------------------------------
>
> Key: MESOS-1420
> URL: https://issues.apache.org/jira/browse/MESOS-1420
> Project: Mesos
> Issue Type: Bug
> Reporter: Robert Lacroix
> Priority: Minor
>
> Slave, framework and run IDs are currently quite long and therefore clutter
> paths to sandboxes. Typically a sandbox path looks like this:
> {code}
> /tmp/mesos/slaves/2014-05-23-16:21:05-16777343-5050-3204-0/frameworks/2014-05-23-16:21:05-16777343-5050-3204-0000/executors/<executor_id>/runs/c22e4dc3-95e5-49fc-9793-b0d22a4f244c
> {code}
> I'd propose shorter and uniform IDs for slaves, frameworks and runs (and
> probably everywhere else where we use IDs) that look like this:
> {code}
> [a-z0-9]{13}
> {code}
> This has about 65bit keyspace compared to 128bit of a UUID, but I think it
> should be random enough.
> With that the path would be roughly 80 chars shorter (179 vs 99) and a lot
> more readable:
> {code}
> /tmp/mesos/slaves/i0b195fb1j14n/frameworks/gtq5kgba60ll4/executors/<executor_id>/runs/7qisjqsb581io
> {code}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
