Merge branch 'main' of ../pcp

qa/1151

@@ -10,13 +10,14 @@ echo "QA output created by $seq"
 
 . ./common.bcc
 
+$sudo rm -rf $tmp.* $seq.full
+
 _pmdabcc_check
 _pmdabcc_require_kernel_version 4 7
 _java_check
 
 status=1       # failure is the default!
 signal=$PCP_BINADM_DIR/pmsignal
-$sudo rm -rf $tmp.* $seq.full
 
 # Launch a Java test program
 cd java

qa/1525

@@ -30,12 +30,18 @@ which pmview >/dev/null 2>&1 || _notrun "No pmview binary installed"
 status=0	# success is the default!
 trap "_cleanup_qt; exit \$status" 0 1 2 3 15
 
+# $tmp.ctrls line is pending a TODO fix for osvis and strange
+# error messages on vm11
+#
 _filter()
 {
     _filter_qt \
     | sed \
 	-e "s@$tmp@TMP@g" \
-	-e '/cannot open/s@/tmp/[0-9][0-9]*[email protected]@' \
+	-e '/cannot open/{
+s@/tmp/[0-9][0-9]*[email protected]@
+s@1: /usr/bin/osvis:@1:@
+}' \
     # end
 }
 

qa/1619

@@ -34,13 +34,16 @@ _filter()
 # real QA test starts here
 inode0=`ls -i $PCP_PMCDCONF_PATH | awk '{ print $1 }'`
 echo Inode0: $inode0 >> $here/$seq.full
+cp $PCP_PMCDCONF_PATH $tmp.save
+cat $PCP_PMCDCONF_PATH >> $here/$seq.full
 
 _service pmcd restart 2>&1 | _filter_pcp_stop | _filter_pcp_start
 _wait_for_pmcd
 _wait_for_pmlogger
 
 inode1=`ls -i $PCP_PMCDCONF_PATH | awk '{ print $1 }'`
 echo Inode1: $inode1 >> $here/$seq.full
+cat $PCP_PMCDCONF_PATH >> $here/$seq.full
 
 echo "Comparing pmcd.conf inode numbers"
 if test $inode0 -eq $inode1 
@@ -51,5 +54,13 @@ else
     status=1
 fi
 
+if diff $tmp.save $PCP_PMCDCONF_PATH
+then
+    :
+else
+    echo "Arrgh ... $PCP_PMCDONF_PATH changed"
+    status=1
+fi
+
 # all done
 exit

qa/admin/list-packages

@@ -13,6 +13,7 @@ _usage()
     echo >&2 "  -c           check packing list"
     echo >&2 "  -i tag       include only lines with a tag"
     echo >&2 "  -m           list only missing packages"
+    echo >&2 "  -n           report my packing-list file and quit"
     echo >&2 "  -v           verbose (debugging)"
     echo >&2 "  -V version   override platform version"
     echo >&2 "  -x tag       exclude lines with a tag"
@@ -264,7 +265,8 @@ include=''
 exclude=''
 missing=false
 check=false
-while getopts 'A:cD:i:mvV:x:?' p
+nflag=false
+while getopts 'A:cD:i:mnvV:x:?' p
 do
     case "$p"
     in
@@ -288,6 +290,9 @@ do
 	m)	missing=true
 		;;
 
+	n)	nflag=true
+		;;
+
 	v)	if $verbose
 		then
 		    very_verbose=true
@@ -406,6 +411,13 @@ else
     exit
 fi
 
+if $nflag
+then
+    echo "$home/package-lists/${distro}+${version}+${arch}"
+    status=0
+    exit
+fi
+
 if [ -n "$exclude" ]
 then
     $very_verbose && [ -n "$exclude" ] && echo >&2 "Excluding tags:$exclude"

qa/admin/package-lists/CentOS+Stream8+x86_64

@@ -36,7 +36,8 @@ grep
 iproute
 httpd
 initscripts
-java-latest-openjdk-headless
+# -latest- (aka 22) and -21- don't have USDT support, so go with -17-
+java-17-openjdk-headless
 libbpf
 libicu-devel
 libpfm-devel

qa/archives/mk.daily_report

@@ -5,6 +5,13 @@
 # a bit of a work out for the pmrep .conf files: sar.conf, vmstat.conf
 # and numa.conf
 #
+# Best to run this on a recent Fedora (e.g. 40 as of Jul 2024) so all
+# of the metrics here are available, and with some concurrent activity
+# for abouth 20 seconds, e.g. I rebuild all the QA applications:
+#   $ cd src/pcp/qa/src
+#   $ make clean
+#   $ make
+#
 
 tmp=/var/tmp/mk-daily-$$
 trap "rm -f $tmp.*; exit 0" 0 1 2 3 15

qa/common.bcc

@@ -211,20 +211,31 @@ _java_check()
 	/etc/alternatives/jre/lib/$__java_machine/server \
 	/usr/lib/jvm/java-${__java_major}-openjdk-$__java_machine/lib/server \
 	/usr/lib/jvm/java-${__java_version}-openjdk/jre/lib/$__java_machine/server \
-	/usr/lib/jvm/java-${__java_major}-openjdk-$__java_machine/jre/lib/$__java_machine/server
+	/usr/lib/jvm/java-${__java_major}-openjdk-$__java_machine/jre/lib/$__java_machine/server \
+	/usr/lib/jvm/jre/lib/server \
+	/usr/lib/jvm/jre-openjdk/lib/server \
+	/usr/lib/jvm/jre-${__java_major}/lib/server \
+	/usr/lib/jvm/jre-${__java_major}-openjdk/lib/server
     do
-        [ -f "$__path/libjvm.$DSO_SUFFIX" ] && __jvm_path="$__path/libjvm.$DSO_SUFFIX"
-    done
-
-    [ -z "$__jvm_path" ] && _notrun "libjvm.$DSO_SUFFIX not found"
-
-    for __tplistcmd in /usr/share/bcc/tools/tplist tplist-bpfcc tplist
-    do
-        which $__tplistcmd >/dev/null 2>&1 && \
-            __probes=$($__tplistcmd -l $__jvm_path 2>/dev/null | grep thread__start)
-        [ -n "$__probes" ] && return 0
+        [ -f "$__path/libjvm.$DSO_SUFFIX" ] || continue
+	__jvm_path="$__path/libjvm.$DSO_SUFFIX"
+	echo "trying \$__jvm_path=$__jvm_path for USDT probes" >>$here/$seq.full
+	for __tplistcmd in /usr/share/bcc/tools/tplist tplist-bpfcc tplist
+	do
+	    which $__tplistcmd >/dev/null 2>&1 && \
+		__probes=$($__tplistcmd -l $__jvm_path 2>/dev/null \
+		           | grep thread \
+			   | tee -a $here/$seq.full \
+			   | grep thread__start)
+	    if [ -n "$__probes" ]
+	    then
+		echo "success with \$__jvm_path=$__jvm_path" >>$here/$seq.full
+		return 0
+	    fi
+	done
     done
-    _notrun "No Java USDT probes available"
+    [ -z "$__jvm_path" ] && _notrun "no libjvm.$DSO_SUFFIX not found"
+    _notrun "No Java USDT probes available in any libjvm.$DSO_SUFFIX"
 }
 
 _python_probe_check()

src/pmdas/linux/help

@@ -211,6 +211,13 @@ of traffic over different paths. See also multipathd(1), where the
 same ID strings (WWID) are used to identify different paths to each
 physical device.
 
+@ hinv.disk.ctlr physical disk device controller
+The disk controller is extracted from /sys/devices/pci0000:00/0000:<ctlr>.
+Multiple disks may share the same controller.
+
+@ hinv.disk.model physical disk device model
+The disk model is extracted from /sys/devices/.../model.
+
 @ disk.dev.avactive per-disk count of active time
 Counts the number of milliseconds for which at least one I/O is in
 progress for each device.

src/pmdas/linux/pmda.c

@@ -18,6 +18,7 @@
  * for more details.
  *
  * pmDebugOption:
+ *   appl1	disk-related metrics
  *   appl8	filesys metrics
  *   libpmda	historic "everything else" diagnostics
  */
@@ -2597,6 +2598,16 @@ static pmdaMetric metrictab[] = {
       { PMDA_PMID(CLUSTER_STAT,104), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_INSTANT, 
       PMDA_PMUNITS(0,0,0,0,0,0) }, },
 
+/* hinv.disk.ctlr */
+    { NULL, 
+      { PMDA_PMID(CLUSTER_STAT,105), PM_TYPE_STRING, DISK_INDOM, PM_SEM_DISCRETE, 
+      PMDA_PMUNITS(0,0,0,0,0,0) }, },
+
+/* hinv.disk.model */
+    { NULL, 
+      { PMDA_PMID(CLUSTER_STAT,106), PM_TYPE_STRING, DISK_INDOM, PM_SEM_DISCRETE, 
+      PMDA_PMUNITS(0,0,0,0,0,0) }, },
+
 /*
  * zram IO cluster
  */

src/pmdas/linux/proc_partitions.c

@@ -828,6 +828,8 @@ static pmID disk_metric_table[] = {
     /* disk.all.flush_rawactive */   PMDA_PMID(CLUSTER_STAT,102),
     /* hinv.map.scsi_id */	     PMDA_PMID(CLUSTER_STAT,103),
     /* disk.all.inflight */	     PMDA_PMID(CLUSTER_STAT,104),
+    /* hinv.disk.ctlr */	     PMDA_PMID(CLUSTER_STAT,105),
+    /* hinv.disk.model */	     PMDA_PMID(CLUSTER_STAT,106),
 
     /* disk.partitions.read */	     PMDA_PMID(CLUSTER_PARTITIONS,0),
     /* disk.partitions.write */	     PMDA_PMID(CLUSTER_PARTITIONS,1),
@@ -1066,6 +1068,138 @@ _pm_ioscheduler(const char *device)
     return "unknown";
 }
 
+/*
+ * Get controller (name) for a specific disk
+ */
+char *
+get_disk_ctlr(char *name)
+{
+    ssize_t		size;
+    int			want;
+    char		*part;
+    char		path[MAXPATHLEN];
+    char		link[MAXPATHLEN];
+    char		*ctlr;
+
+    pmsprintf(path, sizeof(path), "%s/sys/block/%s", linux_statspath, name);
+    if ((size = readlink(path, link, sizeof(link)-1)) < 0) {
+	if (pmDebugOptions.appl1) {
+	    fprintf(stderr, "get_disk_ctlr(%s,...): readlink(%s,...) failed: %" FMT_INT64, name, path, (int64_t)size);
+	    if (size < 0)
+		fprintf(stderr, ": %s", pmErrStr(-oserror()));
+	    fputc('\n', stderr);
+	}
+	return NULL;
+    }
+    link[size] = '\0';
+    /*
+     * .../pci0000:00/0000:00:10.0/.../sdd
+     *                     ^^^^^^^ controlller id as per lspci
+     *                                 ^^^ disk name as per indom
+     */
+    part = strtok(link, "/");
+    want = 0;
+    while (part != NULL) {
+	if (strcmp(part, "pci0000:00") == 0) {
+	    /*
+	     * this is the only PCI prefix we are sure about
+	     * TODO - other possibilities here?
+	     */
+	    want = 1;
+	}
+	else if (want == 1) {
+	    if (strncmp(part, "0000:", 5) == 0) {
+		ctlr = strdup(&part[5]);
+		if (ctlr == NULL)
+		    pmNoMem("get_disk_ctlr: ctlr", strlen(&part[5])+1, PM_RECOV_ERR);
+		return ctlr;
+	    }
+	    else {
+		/* TODO - prefixes other than 0000: here? */
+		if (pmDebugOptions.appl1) {
+		    fprintf(stderr, "get_disk_ctlr(%s,...): expected 0000: got %5.5s from link %s\n", name, part, link);
+		}
+		return NULL;
+	    }
+	}
+	part = strtok(NULL, "/");
+    }
+
+    if (pmDebugOptions.appl1)
+	fprintf(stderr, "get_disk_ctlr(%s,...): link=%s not expected\n", name, link);
+    return NULL;
+}
+
+/*
+ * Get model for a specific disk
+ */
+char *
+get_disk_model(char *name)
+{
+    ssize_t		size;
+    int			fd;
+    char		*part;
+    char		path[MAXPATHLEN];
+    char		link[MAXPATHLEN];
+    char		duplink[MAXPATHLEN];
+    char		*model;
+
+    pmsprintf(path, sizeof(path), "%s/sys/block/%s", linux_statspath, name);
+    if ((size = readlink(path, link, sizeof(link)-1)) < 0) {
+	if (pmDebugOptions.appl1) {
+	    fprintf(stderr, "get_disk_model(%s,...): readlink(%s,...) failed: %" FMT_INT64, name, path, (int64_t)size);
+	    if (size < 0)
+		fprintf(stderr, ": %s", pmErrStr(-oserror()));
+	    fputc('\n', stderr);
+	}
+	return NULL;
+    }
+    link[size] = '\0';
+    strcpy(duplink, link);
+    model = NULL;
+    part = strtok(link, "/");
+    while (part != NULL) {
+	if (strcmp(part, "block") == 0) {
+	    /*
+	     * .../pci0000:00/.../a:b:c:d/block/sdd
+	     * becomes
+	     * .../pci0000:00/.../a:b:c:d/model
+	     */
+	    int		offset;
+	    offset = part - link - 1;
+	    duplink[offset] = '\0';
+	    pmsprintf(path, sizeof(path), "%s/sys/block/%s/model", linux_statspath, duplink);
+	    if ((fd = open(path, O_RDONLY)) >= 0) {
+		char	buf[1024];
+		size = read(fd, buf, sizeof(buf)-1);
+		if (size > 0) {
+		    buf[size-1] = '\0';
+		    model = strdup(buf);
+		    // TODO pmMem
+		    return model;
+		}
+		else {
+		    if (pmDebugOptions.appl1) {
+			fprintf(stderr, "get_disk_model(%s,...): read(%s): %" FMT_INT64 "\n", name, path, (int64_t)size);
+		    }
+		    return NULL;
+		}
+	    }
+	    else {
+		if (pmDebugOptions.appl1) {
+		    fprintf(stderr, "get_disk_model(%s,...): open(%s,...) failed: %s\n", name, path, pmErrStr(-oserror()));
+		}
+		return NULL;
+	    }
+	}
+	part = strtok(NULL, "/");
+    }
+
+    if (pmDebugOptions.appl1)
+	fprintf(stderr, "get_disk_model(%s,...): link=%s not expected\n", name, link);
+    return NULL;
+}
+
 int
 proc_partitions_fetch(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
 {
@@ -1225,6 +1359,20 @@ proc_partitions_fetch(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
 		return PM_ERR_INST;
 	    atom->cp = _pm_scsi_id(p->namebuf);
 	    break;
+	case 105: /* hinv.disk.ctlr */
+	    if (p == NULL)
+		return PM_ERR_INST;
+	    if (p->ctlr == NULL)
+		p->ctlr = get_disk_ctlr(p->namebuf);
+	    atom->cp = p->ctlr;
+	    break;
+	case 106: /* hinv.disk.model */
+	    if (p == NULL)
+		return PM_ERR_INST;
+	    if (p->model == NULL)
+		p->model = get_disk_model(p->namebuf);
+	    atom->cp = p->model;
+	    break;
 	default:
 	    /* disk.all.* is a singular instance domain */
 	    atom->ull = 0;

src/pmdas/linux/proc_partitions.h

@@ -60,6 +60,8 @@ typedef struct {
     char		*dmname;      /* symlink from /dev/mapper, else NULL */
     char		*mdname;      /* symlink from /dev/md, else NULL */
     char		*wwidname;    /* wwid of sd path, else NULL */
+    char		*ctlr;        /* from /sys/block symlink, else NULL */
+    char		*model;       /* from /sys/devices/..., else NULL */
     zram_stat_t		*zram;
     unsigned long long	rd_ios;
     unsigned long long	rd_merges;