[rancid] Cisco devices and calculating percent free

[Emille Blanc]

> ASR920, with churn of whatever in bootflash:/tracelogs

I've been through the ringer on this one (about this time last year, in fact).
If you dig even deeper, it seems to be attached to the cylon_mgr process, which presumably ties to the chipset.
In our environment, busier boxes have greater churn (in the 100's of MBytes/day).
Our patch around this (pre RANCID v3.10) was not to ignore the disk space, but rather impose an upper ceiling for our ASR920's.
Eg;
- !Flash: bootflash: 1339412480 bytes total (470 MB free)
+ !Flash: bootflash: 1339412480 bytes total (>300 MB free)

This way, we still get insight into run-away processes that suddenly burn through the box' flash:, and it adequately stops the noise.
Since upgrading 3.10, it has been less frequent, so we never re-patched.

But can confirm, out of all our devices, it's only the ASR920's that exhibit this behavior in our environment.
ASR901's, 1000's seem fine.

Food for further thought.

-----Original Message-----
From: Rancid-discuss [mailto:rancid-discuss-bounces at www.shrubbery.net] On Behalf Of Chris Wopat
Sent: Friday, November 13, 2020 6:31 AM
To: Randy Bush
Cc: rancid-discuss at www.shrubbery.net
Subject: Re: [rancid] Cisco devices and calculating percent free

Looked into this a little further.

The disk space churn is being caused by this device type - Cisco
ASR920, with churn of whatever in bootflash:/tracelogs . IOS-XE seems
to be doing its own thing in there as files tend to be <1 day old
while no device changes are happening. The contents of this directory
are properly ignored and aren't in diffs, but this slight change in
disk space is causing the issue.

If I do comment this from rancid.types.conf or rancid.types.base, as
expected it removes the directory contents, which does contain
valuable info I like to have (ie which IOS images are chilling there):

#cisco;command;ios::DirSlotN;dir /all bootflash:

So, the only way for me to currently workaround this is to make a
change directly to rancid.pm, in "sub diskszsummary", such as, but not
limited to:

    #return("$total total ($pcnt% free)");
    return("$total total");

Certainly dirty but works. Curious if there's any thoughts on a
'clean' way to do this? I know there's an existing 'FILTER_OSC' flag
that I believe is intended to reduce churn in general, my best guess
for a 'clean' fix, if it makes sense to people, would be to check that
here and only return "$pcnt% free" if that isn't set.

My $.02

If people are opposed to such a thing, I'll likely add this to my
'manually patch these couple of things' list that already exists
post-upgrades.

Cheers
--Chris

_______________________________________________
Rancid-discuss mailing list
Rancid-discuss at www.shrubbery.net
https://www.shrubbery.net/mailman/listinfo/rancid-discuss