Licensing
1) When I run the program using yarn jar, I get an error:
(a)
INFO client.Client: Initializing Client
INFO client.Client: Verifying the license...
java.io.FileNotFoundException: (license file path) (No such file or directory)
at java.io.FileInputStream.open(Native Method)
at java.io.FileInputStream.(FileInputStream.java:146)
This happens when the variable "certificate_file_path", defined in mpi-site.xml, does not point to the License file that is required to run iNFORMER. Please correct the variable to point to the right location and ensure the file has appropriate read permissions as well.
(b)
INFO client.Client: Initializing Client
INFO client.Client: Verifying the license...
(User Information)
Error opening Public Key (Public Key Path)
139877761742496:error:02001002:system library:fopen:No such file or directory:bss_file.c:398:fopen('(path)','r')
139877761742496:error:20074002:BIO routines:FILE_CTRL:system lib:bss_file.c:400:
unable to load Public Key
INFO client.Client: You have 46969620 seconds left
INFO client.Client: License could not be verified. Stop...
This occurs when public_key_filepath, defined in mpi-site.xml does not point to the Public Key file. Please correct the variable to point to the right location and ensure appropriate read permissions as well.
(c)
INFO client.Client: Initializing Client
INFO client.Client: Verifying the license...
(User Data)
FATAL client.Client: Error running CLient
java.lang.NoClassDefFoundError: org/python/util/PythonInterpreter at
(module path).LicenseManager.verifySignature(LicenseManager.java:38)
This occurs when the Jython standalone JAR is not placed in the Yarn application classpath. Please refer to the Getting Started with iNFORMER page on the Jython JAR to download and place it one of the paths mentioned in yarn.application.classpath.
Executing Programs
1) When I submit my code to Yarn, I get the following error:
INFO client.Client: [stdout]: Starting server with 1 clients...
INFO client.Client: [stderr]: [mpiexec@(host)] HYDU_sock_connect (utils/sock/sock.c:172): unable to connect from (host1) to (host2) (Connection refused)
INFO client.Client: [stderr]: [mpiexec@(host)] HYD_pmcd_pmi_publish (pm/pmiserv/pmiserv_pmi.c:117): error connecting to the nameserver
16/07/06 11:05:34 INFO client.Client: [stderr]: [mpiexec@(host)] fn_publish_name (pm/pmiserv/pmiserv_pmi_v1.c:652): error publishing service
This occurs when the Hydra nameserver is not started. In Yarn integration mode, the presence of the Hydra nameserver is necessary for correct functioning of the code. Please start the Hydra nameserver and try again.
In Linux, this can be done by (assuming the Hydra nameserver is in the PATH and in the MPICH installation)
hydra_nameserver &
2) When I try to test my code in standalone mode, using mpiexec, I get the following error:
mpiexec -n 2 (smart binary) (input args)
Fatal error in MPI_Lookup_name: Invalid service name (see MPI_Publish_name), error stack:
MPI_Lookup_name(149): MPI_Lookup_name(service="mpi_yarn_service", MPI_INFO_NULL, port=(memory address)) failed
MPI_Lookup_name(129):
MPID_NS_Lookup(85)..: Lookup failed for service name mpi_yarn_service
To carry out standalone development, Yarn integration must be disabled. This is done by defining the macro DISABLE_YARN_INTEGRATION that must be defined before the first instance of scheduler.h, i.e.
#define DISABLE_YARN_INTEGRATION
#include "histogram.h" // this may include scheduler.h internally.
#include "scheduler.h"
Make sure that you remove the macro and recompile before running through Yarn, else you would notice the following issue.
3) When I run my executable through Yarn, I notice that my job hangs with the following output:
INFO client.Client: [stdout]: Starting server with 1 clients...
INFO client.Client: all containers are launched successfully, executing mpiexec...
INFO client.Client: waiting for service to complete
(Job does not proceed further and seems to hang)
This is related to Issue 2. When running through Yarn, make sure that the DISABLE_YARN_INTEGRATION macro is removed and recompile the binaries to ensure they are up to date, which should resolve this issue.
4) When I run my executable through Yarn, I get the following error:
INFO client.Client: Initializing ApplicationMaster
INFO client.Client: all containers are launched successfully, executing mpiexec...
INFO client.Client: [stderr]: Fatal error in MPI_Lookup_name: Invalid service name (see MPI_Publish_name), error stack:
INFO client.Client: [stderr]: MPI_Lookup_name(149): MPI_Lookup_name(service="mpi_yarn_service", MPI_INFO_NULL, port=(memory address)) failed
INFO client.Client: [stderr]: MPI_Lookup_name(129):
This occurs when CGroups is not enabled as part of the Yarn configuration. Please refer to the User Manual at the Getting Started with iNFORMER page on the relevant Hadoop settings for iNFORMER.
Other Issues
For any issues that are not outlined above, please contact the support team at This email address is being protected from spambots. You need JavaScript enabled to view it..
Thank You.