Package org.acemq.workloads.endurance


package org.acemq.workloads.endurance
  • Class
    Description
    Starts the standing loads, or adopts ones already running, reports on them, and stops them.
    A load being watched: its pid, and whether this call started it.
    A load could not be started or stopped.
    What a load is doing right now.
    The leak drill: standing loads in every language, and every connection torn out from under all of them, over and over, while the run watches what each process is holding rather than what it is publishing.
    The repeated recovery fault, and the connection count that proves it did something.
    What the run is doing, for a console or a terminal.
    What the run found and where it wrote it.
    java -jar acemq-workload.jar endurance [options]: the soak.
    What an endurance run is asked to do: the soak's numbers, the clients to watch and how each one is started, the allowances a leak has to cross, and the upstream defects allowed within a measured ceiling.
    Growth a client is allowed before it is a leak.
    A breach a library is known to have for a reason nothing in the workspace can fix, with a ceiling of its own: base * factor + perRecovery * recoveries + plus.
    How one standing load is started.
    java -jar acemq-workload.jar loads up|down|status: the standing loads a chaos drill watches, started, stopped and reported on through the same ClientLauncher the soak uses.
    What a process is holding, read from outside it.
    Linux: /proc/<pid>/status for VmRSS and Threads, and the entries of /proc/<pid>/fd for descriptors.
    macOS and other Unixes: ps -o rss=, lsof -p and ps -M, counted exactly as scripts/soak.sh counted them so a report from either compares with the other: every line lsof prints, header included, and every thread line ps prints after its header.
     
    One client, read once from outside: one line of the readings TSV.
    The soak's verdict, from its readings: medians of the baseline and final readings per client, against the allowances, with known upstream breaches classified rather than failed.
    One client's before and after.