Hi All,
I am trying setup ET on a new cluster (hosted by the Minnesota Supercomputing Institute at U of Minnesota, Twin Cities). I have been using the mdb file bluewaters.ini as a template, replacing bluewaters specs with my specs where necessary. I've attached my mdb file for reference. I am able to build simfactory using the --machine=mesabi flag. I am able to run the testsuite with no failed tests (although several "unrunnable" tests, which I am assuming is ok?). I can submit jobs, and they seem to run to completion just fine.
However, every time I log in, I am on a different login node, which triggers an "unknown machine name" error, unless I have previously built simfactory on that particular node. I have tried to mimick bluewaters' aliaspattern line in hopes that it would do the trick. But I must be doing something wrong. What do I need to include in my mdb file to force the system to recognize that all of the login nodes are on the same machine?
For a bit more background: MSI uses a two step login process. First you ssh into a login machine. Then you ssh from there into one of the clusters. The machine I eventually reach is named mesabi, and the login hosts are named ln000[1-6].
Any help is greatly appreciated.
Thanks, Eric
Eric
Simfactory's key "aliaspattern" expects a regular expression. This allows you to match several host names, which are then all recognized as being the same machine. For example, Blue Waters has
aliaspattern = ^h2ologin[1-4](.ncsa.illinois.edu)?$
which matches four different host names.
There are two other solutions to this. The one I'm usually using (for other reasons, not because of Simfactory) is to explicitly log in to one of the specific login nodes (e.g. ln0003) after logging in to mesabi. This way Simfactory won't be confused.
If you have a home directory file system that is not shared with other machines, then you can also create a file "$HOME/.hostname" with the content "mesabi" (or similar). Such a file will override Simfactory's host name detection. However, the disadvantage is that each user on this machine will then have to create such a file.
-erik
On Wed, Feb 7, 2018 at 10:29 PM, Eric West ewest@d.umn.edu wrote:
Hi All,
I am trying setup ET on a new cluster (hosted by the Minnesota Supercomputing Institute at U of Minnesota, Twin Cities). I have been using the mdb file bluewaters.ini as a template, replacing bluewaters specs with my specs where necessary. I've attached my mdb file for reference. I am able to build simfactory using the --machine=mesabi flag. I am able to run the testsuite with no failed tests (although several "unrunnable" tests, which I am assuming is ok?). I can submit jobs, and they seem to run to completion just fine.
However, every time I log in, I am on a different login node, which triggers an "unknown machine name" error, unless I have previously built simfactory on that particular node. I have tried to mimick bluewaters' aliaspattern line in hopes that it would do the trick. But I must be doing something wrong. What do I need to include in my mdb file to force the system to recognize that all of the login nodes are on the same machine?
For a bit more background: MSI uses a two step login process. First you ssh into a login machine. Then you ssh from there into one of the clusters. The machine I eventually reach is named mesabi, and the login hosts are named ln000[1-6].
Any help is greatly appreciated.
Thanks, Eric
-- Eric J West Assistant Professor Department of Physics and Astronomy University of Minnesota Duluth
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
On Wed, Feb 07, 2018 at 09:29:16PM -0600, Eric West wrote:
For a bit more background: MSI uses a two step login process. First you ssh into a login machine. Then you ssh from there into one of the clusters. The machine I eventually reach is named mesabi, and the login hosts are named ln000[1-6].
Hi Eric
When taking a look at https://www.msi.umn.edu/content/connecting-hpc-resources
it seems like the login nodes are called 'loginXX' (with X being digits), and from there you ssh into 'mesabi', but end up on nodes with hostnames 'lnXXXX' (with X digits again). This suggests a common file system, in which case you should be able to use a regular expression like ^ln[0-9]{4}$ in 'aliaspattern' (depending on what 'hostname' actually returns on those head nodes; you might have to extend for the fully qualified name).
(Or even more restrictive ^ln000[1-6]$ if you like.)
Frank
Changing the aliaspattern line to either of the following, works:
aliaspattern = ^ln000[1-6]$
aliaspattern = ^ln000[1-6](.msi.umn.edu)?$
The latter corrects a regex mistake on my part (missing '?'). Thanks for the help.
Eric
On 02/08/2018 02:13 AM, Frank Loeffler wrote:
On Wed, Feb 07, 2018 at 09:29:16PM -0600, Eric West wrote:
For a bit more background: MSI uses a two step login process. First you ssh into a login machine. Then you ssh from there into one of the clusters. The machine I eventually reach is named mesabi, and the login hosts are named ln000[1-6].
Hi Eric
When taking a look at https://www.msi.umn.edu/content/connecting-hpc-resources
it seems like the login nodes are called 'loginXX' (with X being digits), and from there you ssh into 'mesabi', but end up on nodes with hostnames 'lnXXXX' (with X digits again). This suggests a common file system, in which case you should be able to use a regular expression like ^ln[0-9]{4}$ in 'aliaspattern' (depending on what 'hostname' actually returns on those head nodes; you might have to extend for the fully qualified name).
(Or even more restrictive ^ln000[1-6]$ if you like.)
Frank
Hello Eric.
If you have a file named ~/.hostname, Simfactory will use that and it will be the same every time. I use that when running in Docker images, etc.
--Steve
On 02/08/2018 02:13 AM, Frank Loeffler wrote:
On Wed, Feb 07, 2018 at 09:29:16PM -0600, Eric West wrote:
For a bit more background: MSI uses a two step login process. First you ssh into a login machine. Then you ssh from there into one of the clusters. The machine I eventually reach is named mesabi, and the login hosts are named ln000[1-6].
Hi Eric
When taking a look at https://www.msi.umn.edu/content/connecting-hpc-resources
it seems like the login nodes are called 'loginXX' (with X being digits), and from there you ssh into 'mesabi', but end up on nodes with hostnames 'lnXXXX' (with X digits again). This suggests a common file system, in which case you should be able to use a regular expression like ^ln[0-9]{4}$ in 'aliaspattern' (depending on what 'hostname' actually returns on those head nodes; you might have to extend for the fully qualified name).
(Or even more restrictive ^ln000[1-6]$ if you like.)
Frank
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
Hello Eric,
if you would like to use simfactory's --remote commands then you can add the first host you have to log in to as a "trampoline" machine in the ini file. Basically you need to have the first host configured as a machine for simfactory (it does not have to be fully functional, just needs to be enough so that you can log into it using "sim login") somewhat similar to how the machine "mars" (in mars.ini) is used by the compute.ini .
Yours, Roland
Hello Eric.
If you have a file named ~/.hostname, Simfactory will use that and it will be the same every time. I use that when running in Docker images, etc.
--Steve
On 02/08/2018 02:13 AM, Frank Loeffler wrote:
On Wed, Feb 07, 2018 at 09:29:16PM -0600, Eric West wrote:
For a bit more background: MSI uses a two step login process. First >> you ssh into a login machine. Then you ssh from there into one of the >> clusters. The machine I eventually reach is named mesabi, and the >> login hosts are named ln000[1-6].
Hi Eric
When taking a look at https://www.msi.umn.edu/content/connecting-hpc-resources
it seems like the login nodes are called 'loginXX' (with X being > digits), and from there you ssh into 'mesabi', but end up on nodes > with hostnames 'lnXXXX' (with X digits again). This suggests a common > file system, in which case you should be able to use a regular > expression like ^ln[0-9]{4}$ in 'aliaspattern' (depending on what > 'hostname' actually returns on those head nodes; you might have to > extend for the fully qualified name).
(Or even more restrictive ^ln000[1-6]$ if you like.)
Frank
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
users@lists.einsteintoolkit.org
-
Eric West -
Erik Schnetter -
Frank Loeffler -
Roland Haas -
Steven R. Brandt