[
https://issues.apache.org/jira/browse/HTRACE-280?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14970309#comment-14970309
]
Colin Patrick McCabe commented on HTRACE-280:
---------------------------------------------
This patch is more about "hooking things up"... i.e. getting the datastore to
track how many spans it dropped because of I/O error or overload, getting the
client to track how many spans were dropped because of buffer overflow, etc.
We already tracked how many spans were added but this patch changes the
mechanism a bit for that.
I agree that it would be nice to get some median, mean, and percentile
calculations. I'm not sure what the best way to do it is since that definitely
involves storing more information. One approach is to do that in a cluster
management system. We could certainly also use a Go framework, or just do the
calculations ourselves based on some stored data.
I also think our GUI should have at least some bare-bones way of getting this
information (currently it's only available via REST and the command-line tool.)
Committed to 4.1...
> htraced: add metrics about total spans added and dropped per address
> --------------------------------------------------------------------
>
> Key: HTRACE-280
> URL: https://issues.apache.org/jira/browse/HTRACE-280
> Project: HTrace
> Issue Type: Bug
> Components: htraced
> Affects Versions: 4.1
> Reporter: Colin Patrick McCabe
> Assignee: Colin Patrick McCabe
> Attachments: HTRACE-280.001.patch, HTRACE-280.002.patch
>
>
> htraced needs metrics about the total spans added and dropped per address.
> This will let us calibrate the sampling rate, and figure out what hosts are
> making the most use of tracing.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
