Am 26.07.2011 um 23:09 schrieb [email protected]: > In the message dated: Tue, 26 Jul 2011 00:30:51 +0200, > The pithy ruminations from Reuti on > <Re: [gridengine users] emails when -M not in script?> were: > => Hi, > => > => Am 26.07.2011 um 00:17 schrieb [email protected]: > => > => > In the message dated: Fri, 22 Jul 2011 21:24:12 +0200, > => > The pithy ruminations from Reuti on > => > <Re: [gridengine users] emails when -M not in script?> were: > => > => Am 22.07.2011 um 21:11 schrieb Mark Suhovecky: > => > => > => > => > Occasionally users submit SGE scripts on our system without the -M > option set. > => > > => > We've got a similar issue. > => > > => > Occasionally users submit SGE scripts to our system with the "-M" > => > option set to an external email address, not accessible from our compute > => > nodes. This causes the node to send multiple "deferred" messages and > => > then a bounce message to the postmaster for each job. It's painful when > => > this happens for 3000 jobs. > => > > => > => > => > => (Second way: Using a JSV to attach an email address to the jobs would > need to look up in a > => datab > => > => ase or the above file, whereto to send the emails. The forwarding > must be set up anyway.) > => > => > => > > => > Execellent timing! One of the items on my to-do list was to send a query > => > about setting up a JSV to deal with invalid email addresses. > => > => I would judge a JSV only of "second class" for this purpose, but personal > taste might vary of co > => urse. What I usually do is to make postfix on the nodes a send-only host, > mangle the sender addr > > Yep, sendmail on our compute nodes is already running to process the queues, > not as a full SMTP daemon. > > => ess to be the one of the master node (as some machine would reject mails > coming from node02@clus > => ter) and use the master this way as a relay. You don't like this solution? > => > > That's fine as a general solution, but the issue at my work is much more > narrow--it's not as much of an issue of configuring the mail servers, as > verifying addresses, which is why I thought a JSV might be more appropriate. > > Most servers here (and all compute nodes) are restricted from sending mail > outside our domain. This is both a data security issue (we're in healthcare) > and an attempt to be good network neighbors (preventing hacked machines from > sending out malware, spam, etc.).
Yes, we have the same setup. Hence the headnode of the cluster contacts the relay of the university to reach the outside world in addition, i.e. the relay on the headnode forwards it to another relay. > Leaving aside questions of the effectiveness of this policy (which I do not > control), my goal is simply to filter out instances where users accidently > enter an external address ([email protected], for example) in their submit > scripts. Optionally, the JSV (or sendmail milter) could mangle those 'invalid' > addresses into the user's internal mail address, based on a lookup of their > username. Yes - okay, then it's a way to go to enforce a policy. In the simplest case it will just remove any email address at all or replace it with the user name or removes it. Do you also check the ~/.forward of the users or ignore them? -- Reuti _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
