Frequently Asked Questions
I get an error relate to file on linux and there is 'too many open files' written somewhere in the error
Which Java version should I use? 🔗OpenMOLE is fully working under OpenJDK 8 and above. OpenJDK is the recommended option. You can check which Java version you're running by typing
java -versionin a console. If you use the closed-source Oracle Java virtual machine (which is probably the case if you are working on Mac or Windows), you have to install the Java Cryptography Extension (JCE) available at the bottom of this page in order for OpenMOLE to be fully functional. JCE is an archive containing a bunch of files that you should replace in the jre/lib/security directory of your java installation. It enables strong cryptographic algorithms. Another concern for Mac users are the concurrent versions of Java that are often present on the same system. Mac OS ships with a default Java 6 that is not suitable to execute OpenMOLE. You must install Java 8 or greater and set the environment variable JAVA_HOME to the install location on your machine (typically the directory containing .../bin/java).
oldVersions 🔗Old versions of OpenMOLE software and documentation are available here. Due to a data loss in 2016, the versions from OpenMOLE 6 are available only.
sshConnectionBug 🔗When one of the SSH authentications you've added to OpenMOLE is marked as failed, you can try these few steps to identify the problems. If you are using OpenMOLE in console mode, enable the FINE level of logging in the console using:
Password authentication} 🔗If you are using the
LoginPasswordauthentication you might want to double check the user and password you entered since one of them is more than likely incorrect.
SSH Keypair AuthenticationIn such a case, we'll have to investigate multiple options, as SSH public key authentications are sensitive to several configuration parameters. Public key authentication has usually a higher priority than password-based authentication when trying to connect to a remote server. Thus, when you attempt an SSH connection to the the target environment, if your client asks you to enter a password (please note that a passphrase is different from a password), then your public key authentication is not taken into account. SSH will skip your public key in case of bad configuration. The most common cases of badly configured keypairs are the following:
- You haven't created an SSH keypair yet (using ssh-keygen). Private keys are usually stored in ~/.ssh/id_rsa or ~/.ssh/id_dsa, and should have a matching ~/.ssh/id_[rd]sa.pub next to them.
- Permissions of your ~/.ssh folder must be set to drwx—— (700 in octal). Also, too permissive home directories (with write access given to the whole group for instance) might show problematic.
- A ~/.ssh/authorized_keys file must be present on the remote system. It should at least contain a line matching the content of the ~/.ssh/id_[rd]sa.pub from your base system.
- You entered a passphrase when you generated your SSH keys and cannot remember it. In such a case, it might be better to generate another keypair.
ssh-keygenshell command. Store it to a different file to avoid overwriting the one already in place. You might also want to try a simple
LoginPasswordauthentication as explained in the SSH h2ion. Adding the
-vvvflag to your ssh command will give a lot more details on the communication between your client and the remote server. This will allow you to find out which authentication is successful as well as the order in which the authentication modes are tried.
Is OpenMOLE doing something? 🔗If you think OpenMOLE is crashed or stuck for some reason, here are a few things you can check from your system to decide whether it's just a temporary slow down or if the platform did actually crash.
Using tools from the Java Development KitA simple call to
jpsfrom your command line will list all the instrumented JVMs on your system. If OpenMOLE is running, it is among these processes. Now that you know OpenMOLE's process id, you can use
jstackto print the eventual stack traces collected from OpenMOLE's threads. It's a bit low level but can at least give you enough material to thoroughly document your problem in the issue list or the mailing list. The same procedure can be applied to the
dbserverrunning along OpenMOLE to manage the replica of the files copied to execution environments.
Inspecting the temporary foldersOpenMOLE automatically creates temporary folders on the machine it's running in order to handle various inputs and outputs. If you have access to the machine running OpenMOLE, change to your OpenMOLE's preferences folder down to the following path: /home/user/.openmole/my_machine/.tmp. List the content of this directory and change to the most recently created directory. If you're using a remote environment, it should contain the tar archives used to populate new jobs on your remote computing environment, along with the input data files required by the task. The only presence of these files is a good indicator that OpenMOLE is functioning correctly and preparing the delegation of your workflow. Hardcore debuggers might want to go even deeper and extract the content of the tar archives to verify them, but this is out of scope. However, touching on temporary file creation in OpenMOLE seamlessly leads us to our next entry...
I've reached my home folder size / file quota 🔗OpenMOLE generates a fair amount of temporary files in the .openmole/mymachine/.tmp folder associated to your machine. Although these are deleted at the end of an execution, they can lead to a significant increase of the space occupied by your .openmole folder and of the number of files present in the same folder. Because some systems place stringent limitations on these two quotas, you might want to move your .openmole folder to a file system not restricted by quotas in order to run your OpenMOLE experiment successfully. The simplest way to do so is to create a destination folder in the unrestricted file system and then create a symbolic link name .openmole in your home directory that points to this newly created folder. On a UNIX system, this procedure translates into the following commands.
In order for this procedure to work, you'll want to ensure the target folder (/data/openmole in the example) can be reached from all the machines running your OpenMOLE installation. Moving your .openmole to a different location is also strongly advised on remote execution hosts (typically clusters) on which you own a personal account used with OpenMOLE. In the case of remote environments, the OpenMOLE runtime and the input files of your workflow will be copied to the .openmole folder, again leading to problematic over quotas on these systems. For this specific case, we recommend using the
# assumes /data is not restricted by quotas cp -r ~/.openmole /data/openmole_data rm -rf ~/.openmole ln -s /data/openmole_data ~/.openmole
sharedDirectoryoption of the cluster environment to set the location where OpenMOLE should copy your files without any hitting quota restrictions.
My sampling generates a type error 🔗Combining samplings is straightforward in OpenMOLE, but can sometimes results in syntax errors a bit cryptic to new users. Let's take the example of a combined sampling made of a file exploration sampling and an integer range exploration:
This combined sampling will generate the following error when compiling the workflow:
(input in (workDirectory / "../data/").files withName inputName) x i in (1 to 10)
OpenMOLE cannot identify the integer range the valid sampling it is. Simply wrapping the expression in parentheses fixes the problem as shown in this correct version:
found : org.openmole.core.workflow.data.Prototype[Int] required: org.openmole.core.workflow.sampling.Sampling
(input in (workDirectory / "../data/").files withName inputName) x (i in (1 to 10))
I can't get CARE / PRoot to work 🔗CARE has recently encountered some compatibility issues with recent versions of the Linux kernel (>= 4.8). If you had downloaded CARE before the release of OpenMOLE 6.0, you might be faced with this issue when trying to package new code. The typical behaviour of this problem is to terminate CARE or your re-execution by a "signal 11" (Segmentation Fault) message. The output will be similar to:
care info: pid 24238: terminated with signal 11. The easiest option to solve this problem is to download a fixed version from GitHub. This binary only works for 64 bits machines, which might not be the case of your workstation. Older versions of CARE are still available for different computer architecture from GitHub. These versions require the
PROOT_NO_SECCOMPenvironment variable to be set to 1 when running CARE. The packaging command line becomes:
PROOT_NO_SECCOMP=1 care -o foo.tgz.bin -r ~ my_app -input 42You can also export this variable to your environment once and for all:
export PROOT_NO_SECCOMP=1You'll have to do that in every terminal you're using unless you set this variable in your shell's startup file (~/.bashrc if you're using bash). There is nothing to change in your OpenMOLE scripts, the good folks at OpenMOLE have everything under control ;) For more information about this specific problem, you can join this discussion.
I get an error relate to file on linux and there is 'too many open files' written somewhere in the error 🔗On linux server the number of file a use can open is generally limited to 1024. OpenMOLE increase it to 4096 when on launch, but it doesn't seem to work on your system, you might want to understand why. To check the current of you system limit execute ulimit -a in a terminal:
In this example you can see the the nax number of open files is 1024. This is generally a soft limitation that can be overridden by the user. To do so execute ulimit -n 4096 before launching in the same terminal. You can check that you command had the expected effect using ulimit -a. If is does't work it means that a hard limit has been set in the limits.conf of your system. If you are administrator you can fix it by modifying the file /etc/security/limits.conf, otherwise you should contact the system administrator and ask him/her kindly to unleash it.
reuillon@docker-host1:~$ ulimit -a core file size (blocks, -c) 0 data seg size (kbytes, -d) unlimited scheduling priority (-e) 0 file size (blocks, -f) unlimited pending signals (-i) 64040 max locked memory (kbytes, -l) 64 max memory size (kbytes, -m) unlimited open files (-n) 1024 pipe size (512 bytes, -p) 8 POSIX message queues (bytes, -q) 819200 real-time priority (-r) 0 stack size (kbytes, -s) 8192 cpu time (seconds, -t) unlimited max user processes (-u) 64040 virtual memory (kbytes, -v) unlimited file locks (-x) unlimited
My workflow fails with QXcbConnection: Could not connect to display 🔗For instance, if you try to use a standalone QGIS application, QGIS might try to access the display server (even when you run the application without a GUI). This will definitely not work on a remote openmole instance (that probably does not have a display server). The full error might look something like:
Fortunately for you, we have a workaround. Let's go back to our QGIS standalone application. It is a python application so we use CARE to embed it in a CARETask in openmole. The packaging would look like:
QXcbConnection: Could not connect to display proot terminated with signal 6
The workaround uses xvfb, a virtual framebuffer X server. The only thing you have to do is to run you command through xvfb. In our example, that would be:
care -o myapp.tgz.bin python3 myapp.py param1 param2
And that's it. After that, no connection to the display server will be attempted.
care -o myapp.tgz.bin xvfb-run python3 myapp.py param1 param2
When shall I use Path over File? 🔗OpenMOLE takes care of everything for you, from connecting to remote environments, to submitting jobs and copying your files. However, most cluster installations take advantage from a shared file system between the nodes. If the file structure you're exploring is located on a shared filesystem, you do not need OpenMOLE to duplicate the target files, as they are available to the compute node directly from the shared filesystem. In case you're manipulating very large files, it might not even be possible to duplicate them. When you find yourself in such a use case, you might want to try the
Pathoptimisation for your scripts. By replacing the
Val[Path]in your scripts, OpenMOLE will store the file's location and not its content as it would when using
Val[File]. This optimisation is only available for cluster and not for the EGI grid. You can find an example of using
Pathvariables in the dataflow in the data processing h2ion.