pythonsql serverUbuntu
I'm running SQL Server 2017 on Ubuntu 16.04.
When trying to run an external Python script, I got the following error:
Msg 39011, Level 16, State 1, Line 0
SQL Server was unable to communicate with the LaunchPad service. Please verify the configuration of the service.
I searched and found out that the solution on Windows is to start the service of LaunchPad with Windows' service manager.
But I need the equivalent of Ubuntu. I don't even know what the LaunchPad service is called.
In order to move TempDB to a disk with more space, I had to restart the server.
This stopped the SQL Server as well as SQL Agent, and the new "SQL Server Launchpad".
It was easy to restart SQL Server and SQL Agent.
To restart LaunchPad, I needed to go to
Administrative Tools -> Services
and find "SQL Server LaunchPad (MSSQLSERVER)"
right-click it and select "Start" and my scripts run again.
UPDATE: It looks like R Services setup is significantly simpler from SQL 2016 RC2 onwards. See here.
This post applies to early CTPs of SQL Server 2016 where additional setup was required for R.
I had this error when I upgraded from CTP3.2 to 3.3. The first thing to check is, if you have a named instance, did you use that name when you ran the post-installation script:
"...rxLibs\x64\registerRext.exe" /install /instance:sql2016
I assume so because you've got the db_rrerole but if not, uninstall using the uninstall switch and try again using your instance name:
"...rxLibs\x64\registerRext.exe" /uninstall
Monitor the output for errors during the install or consider re-directing the output to a file so you can review it afterwards, eg:
"...rxLibs\x64\registerRext.exe" /install /instance:yourNamedInstance > mylog.txt
Check through the log and report any errors you find. If the script succeeds, the Launchpad service should be running and you can check this by running services.msc from the command-line. Here's my service running for the default instance:
Yours should say SQL2016 where mine says MSSQLSERVER. If that's all ok you can open up the the config file, eg from the command-line:
notepad "%programfiles%\Microsoft SQL Server\MSSQL13.MSSQLSERVER\MSSQL\Binn\rlauncher.config"
You will have to change the path above, but the file should be in the Binn directory of your SQL Server install. Inspect the config file and make sure the paths are correct / look sensible. I found initially that there were version numbers on the paths from multiple installs eg ( MSSQLSERVER1000 ) and basically this was interfering with the R install. You can check registry keys HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\MSSQLLaunchpad$SQL2016 and HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Microsoft SQL Server to see if they are consistent but I strongly advise against making changes to your registry, unless you can afford to lose the machine / VM and have to do a full re-install of Windows etc.
I basically did a full uninstall of SQL Server (including a somewhat brutal registry clean-up) and then re-installed a default instance and it worked. I've since resolved not to mix CTPs and set up clean VMs for each one. Also remember CTPs are basically beta products so you should not install them on any servers you value - hopefully this is a test box or non-production VM?
My best advice to you is honestly just use a VM for the CTPs because they are not designed to play well with other versions or even CTP upgrades. I use VirtualBox on my laptop which is fine most of the time. If you have an MSDN subscription (or even a trial one) you can set up a free VM with the latest CTP and throw it away when you're done.
Good luck!
Best Answer
Linux Support Roadmap for SQL Server Machine Learning Services:
"Machine learning using R or Python in-database is not currently supported in SQL Server on Linux. Look for announcements in a later release.
However, on Linux you can perform native scoring using the T-SQL PREDICT function. Native scoring lets you score from a pretrained model very fast, without calling or even requiring an R runtime. This means you can use SQL Server on Linux to generate predictions very fast, to serve client applications."
The stand-alone Machine Learning Server is already supported on Linux.