What does globus endpoint UID not set in .zstash.ini mean?

[forsyth2]

What does globus endpoint UID not set in .zstash.ini mean?

[forsyth2]

Check https://github.com/E3SM-Project/zstash/blob/main/zstash/globus.py for the correct UUID. Alternatively, check on the Globus website directly.

    r"theta.*\.alcf\.anl\.gov": "08925f04-569f-11e7-bef8-22000b9a448b",
    r"blueslogin.*\.lcrc\.anl\.gov": "61f9954c-a4fa-11ea-8f07-0a21f750d19b",
    r"chr.*\.lcrc\.anl\.gov": "61f9954c-a4fa-11ea-8f07-0a21f750d19b",
    r"cori.*\.nersc\.gov": "9d6d99eb-6d04-11e5-ba46-22000b92c6ec",
    r"perlmutter.*\.nersc\.gov": "6bdc7956-fc0f-4ad2-989c-7aa5ee643a79",  # If this doesn't work, use cori

Then:

$ emacs ~/.zstash.ini
[local]
globus_endpoint_uuid = <paste UUID here>

Then, rerun.

[lukaszlacinski]

Zstash checks a fully qualified domain name of the node, zstash is executed on. I do not have access to Perlmutter. What do you get when you run hostname --fqdn?

[forsyth2]

$ hostname --fqdn
login20

[forsyth2]

I know @xylar ran into the issue of Perlmutter's hostname with discover_machine in Mache: https://github.com/E3SM-Project/mache/blob/main/mache/discover.py#L35. Perhaps we could do something similar to that.

[xylar]

On Perlmutter you need to use $NERSC_HOST. If it is unknown, you need to tell users to edit their .bashrc.ext or equivalent to remove the logic that is setting that (a legacy from Hopper and Edison). There seems to be no other good way to identify that you are on Perlmutter.

It seems good to have a command-line flag to identify the host for zstash but the ~/.zstash.ini approach is okay as a workaround.

[forsyth2]

Good to know, thanks!!

[vanroekel]

@lukaszlacinski thanks for the added info. I'm not sure if I fully understand this yet. I initiated the zstash on chrysalis and did not see this error. When I tried to continue on anvil it gave this error. Moving back to chrysalis allowed it to work again. $HOSTNAME gives blueslogin so seems like it should work like on chrysalis as that variable gives chrlogin. Any idea why anvil has this error?

[lukaszlacinski]

When I log into Anvil (blues.lcrc.anl.gov) and check the hostname, I get:

[lukasz@blueslogin1 ~]$ hostname --fqdn
blueslogin1.lcrc.anl.gov`

It matches the regex https://github.com/E3SM-Project/zstash/blob/main/zstash/globus.py#L23.
Can you please run hostname --fqdn when you log in to Anvil?

[vanroekel]

sure no problem -- I get blueslogin3.lcrc.anl.gov but I played with this a bit and think I understand the error. I'm doing the transfer on compute nodes, so if I do the same thing on anvil I get something like b641.lcrc.anl.gov which wouldn't match the regex you link. When I do this on chrysalis I get chr-XXX.lcrc.* which still matches the regex in your link. I think this makes sense now.

[lukaszlacinski]

Temporarily, you can set globus_endpoint.uuid = 61f9954c-a4fa-11ea-8f07-0a21f750d19b in ~/.zstash.ini. I will add r"b\d+.lcrc.anl.gov" to globus.py to match the Blues compute nodes for the next zstash release.

[vanroekel]

thanks for the workaround and for the help understanding this issue!

[forsyth2]

I will add r"b\d+.lcrc.anl.gov" to globus.py to match the Blues compute nodes for the next zstash release.

Created #243 to track that. Thanks!