ATS async log api

[Siddharth Agarwal <sa...@linkedin.com.INVALID>]

Hi,

We are using Logger class defined in the atscppapi. My understanding is
that it flushes logs automatically every 1 second. However, when we see
a crash, we don't get all the logs because they were not flushed and we
lose critical logging information which might have been really helpful in
debugging the crash. We don't want to flush on every write, as it will
impact performance.

Has anybody faced similar issues? Any suggestions on how to get around
this? Has anybody made their logging crash tolerant?

Thanks,
Siddharth Agarwal

Re: ATS async log api

[James Peach <jp...@apache.org>]

> On Aug 16, 2016, at 11:17 AM, Siddharth Agarwal <sa...@linkedin.com.INVALID> wrote:
> 
> Hi,
> 
> We are using Logger class defined in the atscppapi. My understanding is
> that it flushes logs automatically every 1 second. However, when we see
> a crash, we don't get all the logs because they were not flushed and we
> lose critical logging information which might have been really helpful in
> debugging the crash. We don't want to flush on every write, as it will
> impact performance.
> 
> Has anybody faced similar issues? Any suggestions on how to get around
> this? Has anybody made their logging crash tolerant?

One idea is to extend traffic_crashlog to be able to find the log aggregation buffers in a crashing traffic_server process. It could then dump the unflushed logs in a crash report. This might be more mucking around with dwarf than you are interested in, but I expect it could be made to work.

J