• Mac Newbold's avatar
    Several small but much-needed changes to idle detection. Soon I'll · d5bf0bdb
    Mac Newbold authored
    give this stuff a major overhaul when we move to the model where we
    have data on when each node was last "touched" or actively used. (Most
    of these changes will still be relevant.)
    
    1. Fix a bug in idlecheck that we didn't really thing much about. It
    turns out that the WanSpread people have had a 17 node expt idle for
    over three weeks, we didn't detect it because the nodes were running
    ospf and generating lots of network traffic on the exptl. net. We now
    ignore the exptl. network traffic as a source of activity if they have
    automatic ospf routing happening. We also ignore nodes that have any
    trafgen endpoints.
    
    2. After the changes in idlecheck, I updated idle view to correctly
    handle a new flag idlecheck outputs to let us know that an expt has
    ospf running and may be falsely inactive because we ignored net
    traffic. It will probably be very rare that an active expt that is
    running ospf will have only network activity and no tty activity, but
    it is a potential source of false positives. I also did some random
    libifying. There were some hard coded references to emulab.net that I
    fixed to properly use the variables from the defs file.
    
    3. I also updated request_swapexp.php3. It now inlcudes in the email
    message a blurb about automatic swapping, depending on whether or not
    they're marked swappable. (If swappable, it says "This experiment is
    marked as swappable, so it may be automatically swapped out by
    Emulab.Net or its operational staff." and if unswappable, it says
    "This experiment has not been marked swappable, so it will not be
    automatically swapped out.") It also has a reference to the Node Usage
    Policy and gives the URL. So we now give them fair warning about
    potentially getting swapped out and what our policies are.
    d5bf0bdb
idlecheck.in 7.25 KB