[rancid] rancid "losing" characters

Discussion:

Stephen Griffin

2013-10-07 14:30:02 UTC

So, I have a version of rancid 2.3.8 installed, and I've been noticing on
some of our Foundry/Brocade gear that some commands are failing because a
character is lost.. for example, skip-page-dilay instead of
skip-page-display.

Our prior version of rancid was quite old (so old that no one really was
quite sure of what version it was), and we don't recall seeing this
behaviour. It feels like a buffer being exhausted, but 2.3.8 uses
send_human, which should be slowing things down.

The version of expect wasn't changed, and is 5.43.0

We did fold in some of our older custom changes, but I have a difficult
time believing those would impact lost characters.

Has anyone run into this or something similar?

I did a test run, by sending "show 802" to a bunch of devices, and came up
with:
1 e
145 exit
1 sh
1 show 02
1 show 80
145 show 802
1 show802
1 skip-page-ay
148 skip-page-display

Thanks,
Stephen

--
Stephen A. Griffin
Network Engineer
Google Acquisition Netops

heasley

2013-10-16 19:28:30 UTC

Permalink

Post by Stephen Griffin
So, I have a version of rancid 2.3.8 installed, and I've been noticing on
some of our Foundry/Brocade gear that some commands are failing because a
character is lost.. for example, skip-page-dilay instead of
skip-page-display.
Our prior version of rancid was quite old (so old that no one really was
quite sure of what version it was), and we don't recall seeing this
behaviour. It feels like a buffer being exhausted, but 2.3.8 uses
send_human, which should be slowing things down.
The version of expect wasn't changed, and is 5.43.0
We did fold in some of our older custom changes, but I have a difficult
time believing those would impact lost characters.
Has anyone run into this or something similar?

there once was a problem with hpuifilter that caused this, but that would
not affect foundry, only hp/hlogin. perhaps that was part of your changes?

otherwise, i'd suspect expect.

Post by Stephen Griffin
I did a test run, by sending "show 802" to a bunch of devices, and came up
1 e
145 exit
1 sh
1 show 02
1 show 80
145 show 802
1 show802
1 skip-page-ay
148 skip-page-display
Thanks,
Stephen
--
Stephen A. Griffin
Network Engineer
Google Acquisition Netops
_______________________________________________
Rancid-discuss mailing list
http://www.shrubbery.net/mailman/listinfo/rancid-discuss

Gavin McCullagh

2013-10-18 07:48:54 UTC

Permalink

Hi,

I will check this for certain and send on an updated version to the list if
necessary.

Gavin

Post by heasley

there once was a problem with hpuifilter that caused this, but that would
not affect foundry, only hp/hlogin. perhaps that was part of your changes?
otherwise, i'd suspect expect.

Post by Stephen Griffin
I did a test run, by sending "show 802" to a bunch of devices, and came

Post by Stephen Griffin
1 e
145 exit
1 sh
1 show 02
1 show 80
145 show 802
1 show802
1 skip-page-ay
148 skip-page-display
Thanks,
Stephen
--
Stephen A. Griffin
Network Engineer
Google Acquisition Netops
_______________________________________________
Rancid-discuss mailing list
http://www.shrubbery.net/mailman/listinfo/rancid-discuss

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

Gavin McCullagh

2013-10-18 07:49:48 UTC

Permalink

Doh. You can ignore this :-)

Post by Gavin McCullagh
Hi,
I will check this for certain and send on an updated version to the list
if necessary.
Gavin

Post by heasley

Post by Stephen Griffin
So, I have a version of rancid 2.3.8 installed, and I've been noticing

Post by Stephen Griffin
some of our Foundry/Brocade gear that some commands are failing because

Post by Stephen Griffin
character is lost.. for example, skip-page-dilay instead of
skip-page-display.
Our prior version of rancid was quite old (so old that no one really was
quite sure of what version it was), and we don't recall seeing this
behaviour. It feels like a buffer being exhausted, but 2.3.8 uses
send_human, which should be slowing things down.
The version of expect wasn't changed, and is 5.43.0
We did fold in some of our older custom changes, but I have a difficult
time believing those would impact lost characters.
Has anyone run into this or something similar?

there once was a problem with hpuifilter that caused this, but that would
not affect foundry, only hp/hlogin. perhaps that was part of your changes?
otherwise, i'd suspect expect.

Post by Stephen Griffin
I did a test run, by sending "show 802" to a bunch of devices, and came

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

Stephen Griffin

2013-10-18 13:51:29 UTC

Permalink

We had reduced the send_human timers because of internal user complaints
that it was too slow. I went back to the stock value and it dramatically
reduced (although didn't eliminate entirely) the lost characters, so it
appears to be buffer overflows. It is just odd because the older version we
had didn't (ttbomk) use send_human.

Stephen

Post by heasley

there once was a problem with hpuifilter that caused this, but that would
not affect foundry, only hp/hlogin. perhaps that was part of your changes?
otherwise, i'd suspect expect.

Post by Stephen Griffin
I did a test run, by sending "show 802" to a bunch of devices, and came

--
Stephen A. Griffin
Network Engineer
Google Acquisition Netops

heasley

2013-10-18 20:01:42 UTC

Permalink

Post by Stephen Griffin
We had reduced the send_human timers because of internal user complaints
that it was too slow. I went back to the stock value and it dramatically
reduced (although didn't eliminate entirely) the lost characters, so it
appears to be buffer overflows. It is just odd because the older version we
had didn't (ttbomk) use send_human.

sigh. unless there is a bug in tcl/expect, i don't know why that would
happen. when the login scripts get a prompt, they try to build a more
complete prompt for more assurance that it doesnt match stuff in the
output of commands. is it building something reasonable from your prompts
(flogin -d -c somecmd). i do not see anything about buffers in the expect
changelog that seems like it'd affect 5.43. i didnt look at tcl.

Post by Stephen Griffin
Stephen

Post by heasley

there once was a problem with hpuifilter that caused this, but that would
not affect foundry, only hp/hlogin. perhaps that was part of your changes?
otherwise, i'd suspect expect.

Post by Stephen Griffin
I did a test run, by sending "show 802" to a bunch of devices, and came

--
Stephen A. Griffin
Network Engineer
Google Acquisition Netops