Frequently Asked Questions about rancid - last updated 20050813. This FAQ contains information that may not apply directly to versions of rancid prior to 2.3. It also contains paths containing tags such as , which refer to paths that are site-specific and are determined by how rancid is or was configured at installation time. These are explained briefly in the configure --help output. Below are the defaults used in rancid. PREFIX configure --prefix= option. default: /usr/local/rancid EPREFIX configure --exec-prefix= option. default: BINDIR configure --bindir= option. default: /bin The location of clogin, etc. SYSCONFDIR configure --sysconfdir option. default: /etc The location of rancid.conf, etc. LOCALSTATEDIR configure --localstatedir option. default: /var The location of the CVS repository, log files, etc. 1) Platform specific Q. I have a Cisco Catalyst 6500 series switch running the IOS (NOT catOS) software, is the router.db device type cisco or cat5? A. A catalyst running IOS is type "cisco". The 'show version' output will have banner including a phrase similar to "Cisco Internetwork Operating System Software". See the router.db(5) manual page. Q. I have Hybrid Cisco switch, like a cat5k with an RSM. How do I collect both the routing engine and switch configurations? A. Recommended way is to use two entries in the router.db, one for each. For example: cat5k_rsm.domain.com:cisco:up cat5k_sw.domain.com:cat5:up Q. I have a Cisco ??? on which collection stopped working, but clogin works as expected. A. Check if 'write term' produces output. Some IOS combined with large configs and low free memory produce zero 'write term' output, esp. combined with a memory leak. The device will have to be rebooted and/or upgraded. Q. I have a Cisco Catalyst switch. clogin connects, but after receiving the prompt, it stalls until it times out. Why? A. This may be due to your prompt. CatOS does not include an implicit '>' in it's prompt, like IOS does. clogin looks for '>' during login, so specify your prompt with a trailing '>'. Also see cat5rancid(1). For example: cat5k> cat5k> enable Password: cat5k> (enable) Q. Polling a ZebOS box fails from cron, but is successful from the command- line. A. This is the tty/pty handling of either your O/S or ZebOS. Supposedly, changing the TERM in /rancid.conf to the following seems to fix it. TERM=vt100;export TERM COLUMNS=160; LINES=48; export COLUMNS LINES 2) CVS and filesystem permissions WARNING: Be careful when mucking around with the repository! Q. I am new to CVS, where can I find additional information? A. The manual page for CVS is quite complete, but can be be overwhelming even for someone familiar with RCS. There are some excellent resources on the web. See http://en.wikipedia.org/wiki/Concurrent_Versions_System and http://cvshome.org/. Q. Errors are showing up in the logs like: cvs [diff aborted]: there is no version here; run 'cvs checkout' first A. The directory was not imported into CVS properly or was not properly checked out afterward, so CVS control files or directories do not exist. rancid-cvs should always be used to create the directories and perform the CVS work. If it is just the directories that have been created manually, save a copy of the router.db file, then remove the group's directory, use rancid-cvs, and replace the router.db file. If the CVS import was also performed manually, cd to and use 'cvs co ' to create all the CVS control bits. Q. I keep receiving the same diff for a (or set of) devices, but I know the data is not changing repeatedly. Why? A. This is probably a CVS or filesystem permissions problem. Check the log file from the last run for that group for clues first; it may provide the exact cause. Note: It is very important the following be done as the user who normally runs the rancid collection from cron. Check the cvs status of the device's file. example: guelah [2704] cvs status rtr.shrubbery.net =================================================================== File: yogi.shrubbery.net Status: Up-to-date Working revision: 1.197 Tue Jul 10 15:41:16 2001 Repository revision: 1.197 /usr/local/rancid/var/CVS/shrubbery/configs/rtr.shrubbery.net,v Sticky Tag: (none) Sticky Date: (none) Sticky Options: (none) The Status: should be Up-to-date. If the status is "Unknown", then somehow the file has been created without being cvs add'ed. This should be corrected by removing that device's entry from the group's router.db file, run rancid-run, replace the entry in router.db, and run rancid-run again. If the Status is anything else, someone has most likely been touching the files manually. Sane state can be achieved by removing the file and running cvs update to get a fresh copy from the repository. Check the ownership and permissions of the file and directory and the directory and file in the cvs repository (/CVS/). They should be owned by the user who runs rancid-run from cron. At the very least, the directory and files should be writable by the rancid user. Group and world permissions will determined by the umask (default 027), which is set in /rancid.conf. Likely the easiest way to fix the ownership on the cvs repository is chown -R /CVS / Q. I am renaming a device but would like to retain the history in CVS. How is this done? A. CVS does not provide a way (AFAIK) to rename files or to rename or delete directories. The best way is to copy the CVS repository file manually like this (disclaimer: BE VERY CAREFUL mucking around with the repository): % su - rancid_user % cd % echo "new_device_name:device_type:up" >> /router.db % cp -p CVS//configs/old_device_name,v \ CVS//configs/new_device_name,v % cd /configs % cvs update where GROUP is the name of the rancid group that the device is a member of. Once the renaming is complete, remove the old name from the router.db file and leave the CVS clean-up of the old filename to rancid. If one wanted to move a device to a different group and maintain the history, the same procedure would work, substituting the new group name appropriately and editing the router.db of both old and new groups, of course. SVN provides a rename function; but we suggest that you use it's copy function instead, and leave the clean-up of the old name to rancid. So, you would use the copy function (proper substitutions, of course) in place of the cp in the CVS example, then commit the new file. Q. I am removing a group and would like to remove all traces of it from the rancid directory and the CVS repository. How is this done? A. As far as I know, CVS does not provide a way to remove directories. First, remove the group from /rancid.conf. If rancid is running, wait for it to complete. Then just recursively remove the directory. For example, a group named "fubar": % su - rancid_user % cd % rm -rf fubar CVS/fubar Q. I would like to place my CVS repository on a remote machine. How do I do that? A. Assuming that you're starting fresh, its quite simple. Before running rancid-cvs for the first time, adjust CVS_RSH & CVSROOT in rancid.conf similar to the following: CVS_RSH=ssh; export CVS_RSH CVSROOT="myhost:/fqpn/CVS"; export CVSROOT Note that CVS_RSH is not found in the sample rancid.conf that is distributed with rancid. Q. I need a web interface to the rancid CVS repository, for the CVS unsavvy. A. cvsweb works with rancid. Other similar software may as well. http://www.freebsd.org/projects/cvsweb.html cvsweb.conf: @CVSrepositories = ( 'rancid' => ['RANCID CVS, '/full_path_to_the_RANCID_CVS'], where the path will be /CVS. 3) General Q. I have a (set of) device(s) on which collection fails. How can I debug this? A. Our usual diagnostic procedure for this is: - Make sure that the appropriate *login (example: clogin for cisco) works. This tests to make sure you don't have routing or firewall issues, DNS or hostname errors, that your .cloginrc is correct, your banner does not have some character that *login does not like, and that the *login script doesn't have a bug of some sort. For example: clogin cisco_router Should login to cisco_router and produce a router prompt that you can use normally, as if clogin were not used (i.e.: telnet cisco_router). - See if commands can be executed on the router via clogin. This will exercise the *login functionality needed for rancid. For example: clogin -c 'show version; show diag' cisco_router Should login to cisco_router, run show version and show diag, then disconnect and exit. The output will be displayed on your terminal. - Then see if the correct rancid commands work against the router. For example: rancid cisco_router Should produce a cisco_router.new file (cooked to a golden rancid-style colour) in the current directory. If it does not, try again with the -d option, so that the cisco_router.new file will not be removed if an error is detected. Note: if you have NOPIPE set in your environment, a cisco_router.raw file will be produced that is the raw output of the dialogue with the device. If all of these work, make sure that the device's entry in the group's router.db file is correct and check the group's last log file for errors. Q. Are there any characters in the banner that rancid has problems with OR I changed the device's command prompt and now collection is failing? A. The trickiest part about clogin (et al) is recognizing the prompt correctly. clogin looks for '>' and '#' to figure out if it is logged in or in enable mode. So if you have a '>' or '#' in your login banner (or other motd), then clogin gets all confused and will not be able to log in correctly, and thus rancid will fail. Don't use '>' or '#' in your prompt or in your banner or other motd. Q. I use /*login -c to run commands on multiple boxes. Sometimes these are commands that take secondary input, like a filename. How can I enter the data for that secondary prompt? A. Two methods will work. Write an expect script to be used with clogin's -s option, for which a few examples come with rancid like cisco-load.exp. OR provide all the input in one command with the -c option like so: Router#clear counters Clear "show interface" counters on all interfaces [confirm] Router# clogin -c 'clear counters\n' The specific return (\n) will be entered after 'clear counters' followed by the normal return after the command. Some devices apparently eat the linefeed of the typical Unix \r\n sequence and require that a carriage- return be used instead (\r). Q. I would like to collect device configurations every hour, but only receive diffs every Nth collection or every N hours. Is this possible? A. Certainly, but rancid does not provide such a mechanism natively. Two approaches are recommended: 1) Using your preferred mail-list software, add a list with a digest and configure your MTA (example: sendmail) to send diffs to the list. Configure the mail-list software to force the digest at the interval desired. This allows folks to choose which type they prefer, after each collection or every N hours. This method also provides easy methods to archive the diff mail and retrieve previous diffs. 2) Write a script to send diffs, which saves the time it last ran and passes this to the -D option of CVS. Obviously, the first option is the cleanest and most featureful, which is why the script mentioned in the second option is not provided. Q. I'd like to have RANCID automatically begin collection when someone finishes configuring a router. How can I do this? A. Using a syslog watcher script, one can trigger RANCID from the syslog line emitted by, for example, an IOS router after configuration mode is ended. Here's a simple example using the Simple Event Correlator: (http://simple-evcorr.sourceforge.net/) If the syslog line in your logs looks like this (wrapped for readability): Apr 5 09:56:52 acc1.geo269.example.com 72: 000069: *Mar 6 21:40:13.466 \ AEDT: %SYS-5-CONFIG_I: Configured from console by gwbush on vty0 (10.1.1.1) You would use a SEC configuration stanza like this: # example rancid trigger # type=SingleWithSuppress ptype=RegExp pattern=\s\S+:\S+\S+\s(\S+)\.example\.com.*SYS-5-CONFIG_I action=shellcmd /opt/rancid/bin/do-diffs -r $1 window=1800 This will execute the command '/opt/rancid/bin/do-diffs -r acc1.geo269' when it is fed a line like that syslog line. The command will be run at most once every 1800 seconds. If you do not get hostnames in your log lines that match your router.db entries, either fix your reverse DNS or remove the '-r $1' part. Q. I would like to limit the permissions of the rancid user on my devices. Is this possible? A. Strictly speaking, no. Rancid needs permission to read device configuration and other data which is often not available to underprivileged users. However, if you use TACACS+, you can limit the commands that are available to a user. For example, to allow ping and show, but not "show tcp", and nothing else: user = rancid { cmd = "ping" { permit .* } cmd = "show" { deny tcp.* permit .* } # the default is to deny other commands } For RADIUS, Justin Grote suggested privilege levels: http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/122newft/122t/122t13/ftprienh.htm Q. For approximately X hosts (configs) what size server should we be considering - speed and data storage? A. On modern machines it is unlikely you will have issues with disk space or memory - A heavily laden access router with a complex config won't consume more than a few megabytes of disk space for its configs over several years time (roughly 3 times the sum of all the config or */configs/* over 2 years is a decent approximation). Rancid is typically CPU bound if you have adequate network bandwidth. Experience shows rancid takes around 50 Mhz * minutes / device of processing power. This means that a 1Ghz machine can poll: 1000 Mhz * 60 (min/hour) / 50(Mhz min / device) = 1200 devices/hour That's obviously a ball park estimate which varies with many different factors such as the CPU type and the types of devices on your network. Q. How can I run rancid to make the most efficient use of resources (i.e. run in the shortest amount of time)? A. You can adjust PAR_COUNT in rancid.conf to achieve maximum efficiency during polling. You can watch the output of the standard unix command vmstat command during polling to determine whether or not the cpu is being wholly utilized - there should be little idle time and no process blocking (see vmstat). Another simpler method is to look at the time stamps on the rancid log files, and adjust PAR_COUNT until the least amount of time is taken during polling. Make sure all devices are being polled by rancid before using this method - failing devices can extend the amount of time rancid takes to finish by a *LONG* period and throw your times way off. It may help to run rancid niced (man nice) if it will be sharing resources with other processes, as it may eat whatever is available if PAR_COUNT is set high. This is done by changing the crontab to be something like: 5 * * * * nice -19 /usr/local/rancid/bin/rancid-run If you _do_ share resources with other processes but want rancid to run efficiently, probably the vmstat method above will work better - rancid may take a little longer to run but you won't be stepping on other people's toes. Q. I'm still stuck on this problem. Where can I get more help? A. A discussion list is available, rancid-discuss@shrubbery.net. You must be a subscriber to post. Subscribe like this: shell% echo "subscribe" | mail rancid-discuss-request@shrubbery.net Q. What else can I do with rancid? A. The possibilities are endless...rancid is non-toxic when applied properly. see Joe Abley and Stephen Stuart's NANOG presentation: http://www.nanog.org/mtg-0210/abley.html or our NANOG presentation: http://www.shrubbery.net/rancid/NANOG29/