|
I suggested the full slot name as a way to bypass the code that tries to build a fully qualified hostname when you give condor_drain a bare hostname.
When you pass a bare hostname (one without any . characters) to most of the condor commands, they will try and turn that into a fully qualified hostname and then look that up in the collector to get the address to send the command to.
HTCondor has a strong bias towards using full hostnames, but since the ads in your collector do not have fully qualified names when tools convert a name on the command line to a fully qualfied name, that lookup in the collector will not match.
But when you pass a slot name to tools like condor_drain, it will not try and convert that to fully qualified and so that lookup succeeds.
What the HTCondor tools expect you to do to fix this is to configure DEFAULT_DOMAIN_NAME with a consistent value across your pool, or to configure DNS to return fully qualified names across your pool.
-tj
From: Curtis Spencer <curtis.spencer@xxxxxxxxxxxx>
Sent: Monday, June 2, 2025 3:13 PM
To: John M Knoeller <johnkn@xxxxxxxxxxx>
Cc: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] condor_drain error
Using "slot1@condor-execute" worked.
I verified that it actually drained the node and that all jobs were marked as idle until I undrained the machine.
Seems like specifying the slot is not typical since it wasn't your first suggestion. AFAIK, I'm only running the one startd on the machine so it would be nice to be able to not have to specify "slot1@". Is there a config change
that will make that work?
I'm not sure why you would not be seeing more D_HOSTNAME messages.
Try slot1@condor-execute instead of just condor-execute.
I don't get any additional information with the additional debug flags. Here's the output with and without DEFAULT_DOMAIN_NAME set on both condor-master and condor-execute:
root@condor-master:~# condor_drain -debug:D_HOSTNAME,D_FULLDEBUG condor-execute
05/30/25 22:24:55 Can't find address for startd condor-execute.ltc
ERROR: Can't find address for startd condor-execute.ltc
root@condor-master:~# condor_drain -debug:D_HOSTNAME,D_FULLDEBUG condor-execute
ERROR: unknown host condor-execute
Thanks,
Curtis
you were running
condor_drain -debug:D_HOSTNAME,D_FULLDEBUG before.
D_HOSTNAME is nessary to see the messages I mentioned.
The output of `condor_drain -debug condor-execute` (without adding DEFAULT_DOMAIN_NAME) is just "ERROR: unknown host condor-execute".
With "DEFAULT_DOMAIN_NAME = ltc" set on both the master and execute machines, `condor_status` doesn't return anything and the output of `condor_drain -debug condor-execute` and `condor_drain -debug condor-execute.ltc` is:
05/30/25 19:09:17 Can't find address for startd condor-execute.ltc
ERROR: Can't find address for startd condor-execute.ltc
Thanks,
Curtis
You might be able to fix it by adding
DEFAULT_DOMAIN_NAME=
o your configuration.
You have not said what the -debug output of condor_drain is. I would expect to see messages like.
05/29/25 13:21:17.138 Finding proper daemon name for "condor-execute"
05/29/25 13:21:17.138 Daemon name contains no '@', treating as a regular hostname
then it will either print
05/29/25 13:21:17.138 ipv6_getaddrinfo() could not look up condor-execute: ....
or it will append the default domain name to it and print or
05/29/25 13:21:17.138 Returning daemon name: "condor-execute.<default-domain>"
or
05/29/25 13:21:17.138 Failed to construct daemon name, returning NULL.
-tj
What steps can I take to fix that? I'm using the
Execute Role on the machine I am trying to drain and I am able to run jobs on it so I'm confused about why the collector wouldn't be able to find a STARTD with the name of that machine name in the Collector.
condor_status shows the `condor-execute` machine:
root@condor-master:~# condor_status
Name OpSys Arch State Activity LoadAv Mem ActvtyTime
slot1@condor-execute LINUX X86_64 Unclaimed Idle 0.000 3927 0+00:00:00
Total Owner Claimed Unclaimed Matched Preempting Drain Backfill BkIdle
X86_64/LINUX 1 0 0 1 0 0 0 0 0
Total 1 0 0 1 0 0 0 0 0
I've tried explicitly setting the MACHINE ClassAd to "condor-execute" and have restarted Condor but I get the same error from `condor_drain`.
Thanks,
Curtis
Yes, that message indicates that the condor_drain command is not able to get the find a STARTD with that name in the collector when it goes to look up the address of the STARTD.
-tj
I'm still unable to run condor_drain using the machine alias. Any suggestions?
Does that indicate that the failure is from the attempt to get the address from the collector or later on? What else can I do to debug this?
Running `condor_drain -debug:D_COMMAND,D_HOSTNAME,D_FULLDEBUG "condor-execute"` results in the same error: `ERROR: unknown host condor-execute`.
We should try and figure out where the failure is happening.
try
condor_drain -debug:D_COMMAND,D_HOSTNAME,D_FULLDEBUG "condor-execute"
To see if the failure is from the attempt to get the address from the collector, or if it is failing later on.
-tj
From: Curtis Spencer
Sent: Friday, April 18, 2025 3:59 PM
To: John M Knoeller
Cc: HTCondor-Users Mail List
Subject: Re: [HTCondor-users] condor_drain error
Thanks. Running the command with the full address worked:
Running the command using the machine alias returns "ERROR: unknown host condor-execute"
condor_drain -debug "condor-execute"
What can I do to get the machine alias working? That's much easier to use/remember than the full address.
Thanks,
Curtis
The <> around the address and the sock= are part of the address, and are needed.
condor_drain -debug "condor-execute"
should also work, since "condor-execute" is the value of the Machine attribute in the collector.
-tj
From: Curtis Spencer
Sent: Friday, April 18, 2025 12:58 PM
To: John M Knoeller
Cc: HTCondor-Users Mail List
Subject: Re: [HTCondor-users] condor_drain error
I ran that command and tried running `condor_drain` using the entire value of MyAddress but it hung and I had to exit. I also tried with the Ip address and machine alias but got the same errors as before.
```
$ sudo condor_drain -debug 192.168.9.161
04/18/25 17:38:45 Can't find address for startd 192.168.9.161
ERROR: Can't find address for startd 192.168.9.161
$ sudo condor_drain -debug condor-execute
ERROR: unknown host condor-execute
```
Thanks,
Curtis
The HTCondor tools mostly don't use DNS to locate daemons when they want to send commands.
The way most HTCondor tools, including condor_drain, determine the address of a daemon is by querying the collector, so the argument you pass must be the name of the daemon in the collector.
Try running
condor_status -af:h Name Machine MyAddress
The MyAddress value is what condor_drain needs to lookup in order to see send the drain command, It will look for entries in the collector where either the Name or the Machine value matches what you passed to condor_drain.
-tj
I have a test condor pool with two nodes (installed using `get_htcondor`):
* Node A (hostname `condor-master`; IP 192.168.9.7) has `use role:get_htcondor_central_manager` and `use role:get_htcondor_submit` roles
* Node B (hostname `condor-execute`; IP 192.168.9.161) has `use role:get_htcondor_execute`
I ran `systemctl status condor` on both nodes to confirm that the master has condor_collector, condor_negotiator, and condor_schedd running and that the execute node has condor_startd running.
Both nodes have `use security:recommended` in `/etc/condor/config.d/00-security`.
I have run `condor_status` and confirmed that the execute machine has joined the pool and I have run `condor_submit` with a test script and confirmed that the job ran.
I have copied the `/etc/condor/passwords.d/POOL` file from the master node to the execute node and confirmed 0600 permissions for that file on both nodes.
I have run `condor_token_request` on the master node and approved the request on the execute node using `condor_token_request_approve` and stored the returned key in `/etc/condor/tokens.d/admin@condor` on the master node.
I then ran `condor_drain -debug 192.168.9.161` but got this error: "Can't find address for startd 192.168.9.161". I thought this could be related to using the IP address instead of a DNS name so I added an entry to my /etc/hosts
file and ran `condor_drain -debu condor-execute` but I got this error: "ERROR: unknown host condor-execute"
Conceptually, I feel like I understand how IDTOKEN auth is supposed to work (and to some extent, I think it is working since the execute machine was able to join the pool), but I can't figure out why `condor_drain` won't work.
Thanks,
Curtis
|
![[Computer Systems Lab]](http://www.cs.wisc.edu/pics/csl_logo.gif){kind=logo}