reclaim_retry_trace.bt

gistfile1.txt

/*
 * reclaim_retry_trace.bt — Capture should_reclaim_retry() behaviour
 *
 * Uses the built-in reclaim_retry_zone tracepoint to log every
 * retry decision in direct reclaim.  
 *
 * Tracepoint fields:
 *   args.node, args.zone_idx, args.order,
 *   args.reclaimable, args.available, args.min_wmark,
 *   args.no_progress_loops, args.wmark_check
 *
 * no_progress_loops == 0 means did_some_progress was true (reclaim
 * freed at least one page, resetting the counter).  With zram this
 * includes anon pages swapped out synchronously — even though no
 * net free memory was produced.
 *
 * Run:  sudo bpftrace reclaim_retry_trace.bt
 * Stop: Ctrl-C (prints summary histograms)
 */

BEGIN
{
	printf("Tracing should_reclaim_retry via reclaim_retry_zone...\n");
	printf("Hit Ctrl-C for summary.\n\n");
	printf("%-16s %6s %5s %5s %14s %14s %14s %5s %5s\n",
		"COMM", "PID", "NODE", "ORDER",
		"RECLAIMABLE", "AVAILABLE", "MIN_WMARK", "LOOPS", "WMARK");
}

tracepoint:oom:reclaim_retry_zone
{
	printf("%-16s %6d %5d %5d %14lu %14lu %14lu %5d %5d\n",
		comm, pid,
		args.node, args.order,
		args.reclaimable, args.available, args.min_wmark,
		args.no_progress_loops, args.wmark_check);

	@loops_hist = lhist(args.no_progress_loops, 0, 20, 1);

	if (args.wmark_check) {
		@wmark["pass"] = count();
	} else {
		@wmark["fail"] = count();
	}

	@reclaimable = hist(args.reclaimable);
	@total = count();
}

interval:s:5
{
	printf("\n--- 5s snapshot ---\n");
	printf("no_progress_loops distribution:\n");
	print(@loops_hist);
	printf("wmark outcomes:\n");
	print(@wmark);
}

END
{
	printf("\n\n=== FINAL SUMMARY ===\n");
	printf("\nno_progress_loops distribution:\n");
	print(@loops_hist);
	printf("\nreclaimable pages distribution:\n");
	print(@reclaimable);
	printf("\nwmark_check outcomes:\n");
	print(@wmark);
	printf("\ntotal events: ");
	print(@total);

	clear(@loops_hist);
	clear(@reclaimable);
	clear(@wmark);
	clear(@total);
}