* 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*
* Authors:
- * Aurelien Reynaud <collectd at wattapower.net>
+ * Aurélien Reynaud <collectd at wattapower.net>
**/
#include "collectd.h"
#include <libperfstat.h>
#include <sys/utsname.h>
+/* XINTFRAC was defined in libperfstat.h somewhere between AIX 5.3 and 6.1 */
+#ifndef XINTFRAC
+# include <sys/systemcfg.h>
+# define XINTFRAC ((double)(_system_configuration.Xint) / \
+ (double)(_system_configuration.Xfrac))
+#endif
+
+#define CLOCKTICKS_TO_TICKS(cticks) ((cticks) / XINTFRAC)
+
static const char *config_keys[] =
{
"CpuPoolStats",
static _Bool pool_stats = 0;
static _Bool report_by_serial = 0;
+#if PERFSTAT_SUPPORTS_DONATION
+static _Bool donate_flag = 0;
+#endif
+static char serial[SYS_NMLN];
+
+static perfstat_partition_total_t lparstats_old;
static int lpar_config (const char *key, const char *value)
{
return (0);
} /* int lpar_config */
-static void lpar_submit (const char *type_instance, counter_t value)
+static int lpar_init (void)
+{
+ int status;
+
+ /* Retrieve the initial metrics. Returns the number of structures filled. */
+ status = perfstat_partition_total (/* name = */ NULL, /* (must be NULL) */
+ &lparstats_old, sizeof (perfstat_partition_total_t),
+ /* number = */ 1 /* (must be 1) */);
+ if (status != 1)
+ {
+ char errbuf[1024];
+ ERROR ("lpar plugin: perfstat_partition_total failed: %s (%i)",
+ sstrerror (errno, errbuf, sizeof (errbuf)),
+ status);
+ return (-1);
+ }
+
+#if PERFSTAT_SUPPORTS_DONATION
+ if (!lparstats_old.type.b.shared_enabled
+ && lparstats_old.type.b.donate_enabled)
+ {
+ donate_flag = 1;
+ }
+#endif
+
+ if (pool_stats && !lparstats_old.type.b.pool_util_authority)
+ {
+ WARNING ("lpar plugin: This partition does not have pool authority. "
+ "Disabling CPU pool statistics collection.");
+ pool_stats = 0;
+ }
+
+ return (0);
+} /* int lpar_init */
+
+static void lpar_submit (const char *type_instance, double value)
{
value_t values[1];
value_list_t vl = VALUE_LIST_INIT;
- /* Although it appears as a double, value is really a (scaled) counter,
- expressed in CPU x seconds. At high collection rates (< 1 min), its
- integer part is very small and the resulting graphs get blocky. We regain
- some precision by applying a x100 factor before casting it to a counter,
- turning the final value into CPU units instead of CPUs. */
- values[0].counter = value;
+ values[0].gauge = (gauge_t)value;
vl.values = values;
vl.values_len = 1;
-
- /* An LPAR has the same serial number as the physical system it is currently
- running on. It is a convenient way of tracking LPARs as they are moved
- from chassis to chassis through Live Partition Mobility (LPM). */
if (report_by_serial)
{
- struct utsname name;
- if (uname (&name) != 0)
- {
- ERROR ("lpar plugin: uname failed.");
- return;
- }
- sstrncpy (vl.host, name.machine, sizeof (vl.host));
+ sstrncpy (vl.host, serial, sizeof (vl.host));
sstrncpy (vl.plugin_instance, hostname_g, sizeof (vl.plugin));
}
else
sstrncpy (vl.host, hostname_g, sizeof (vl.host));
}
sstrncpy (vl.plugin, "lpar", sizeof (vl.plugin));
- sstrncpy (vl.type, "cpu", sizeof (vl.type));
+ sstrncpy (vl.type, "vcpu", sizeof (vl.type));
sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance));
plugin_dispatch_values (&vl);
-}
-
-static int lpar_read_shared_partition (const perfstat_partition_total_t *data)
-{
- static counter_t time_old;
- static counter_t user_old;
- static counter_t syst_old;
- static counter_t wait_old;
- static counter_t idle_old;
- static counter_t unav_old;
-
- counter_t user = (counter_t) data->puser;
- counter_t syst = (counter_t) data->psys;
- counter_t wait = (counter_t) data->pwait;
- counter_t idle = (counter_t) data->pidle;
- counter_t unav = 0;
-
- /*
- * On a shared partition, we're "entitled" to a certain amount of
- * processing power, for example 250/100 of a physical CPU. Processing
- * capacity not used by the partition may be assigned to a different
- * partition by the hypervisor, so "idle" is hopefully a very small
- * number.
- *
- * We calculate the amount of ticks assigned to a different partition
- * from the number of ticks we're entitled to and the number of ticks
- * we used up.
- */
- if (time_old != 0)
- {
- counter_t time_diff;
- counter_t entitled_ticks;
- counter_t consumed_ticks;
- counter_t user_diff;
- counter_t syst_diff;
- counter_t wait_diff;
- counter_t idle_diff;
- counter_t unav_diff;
-
- double entitled_pool_capacity;
-
- /* Number of ticks since we last run. */
- time_diff = ((counter_t) data->timebase_last) - time_old;
-
- /* entitled_pool_capacity is in 1/100th of a CPU */
- entitled_pool_capacity = 0.01 * ((double) data->entitled_pool_capacity);
-
- /* The number of ticks this partition would have been entitled to. */
- entitled_ticks = (counter_t) ((entitled_pool_capacity * ((double) time_diff)) + .5);
-
- /* The number of ticks actually spent in the various states */
- user_diff = user - user_old;
- syst_diff = syst - syst_old;
- wait_diff = wait - wait_old;
- idle_diff = idle - idle_old;
- consumed_ticks = user_diff + syst_diff + wait_diff + idle_diff;
-
- /* "uncapped" partitions are allowed to consume more ticks than
- * they are entitled to. */
- if (entitled_ticks >= consumed_ticks)
- unav_diff = entitled_ticks - consumed_ticks;
- else
- unav_diff = 0;
- unav = unav_old + unav_diff;
-
- lpar_submit ("user", user);
- lpar_submit ("system", syst);
- lpar_submit ("wait", wait);
- lpar_submit ("idle", idle);
- lpar_submit ("unavailable", unav);
- }
-
- time_old = (counter_t) data->timebase_last;
- user_old = user;
- syst_old = syst;
- wait_old = wait;
- idle_old = idle;
- unav_old = unav;
-
- return (0);
-} /* int lpar_read_shared_partition */
-
-static int lpar_read_dedicated_partition (const perfstat_partition_total_t *data)
-{
- lpar_submit ("user", (counter_t) data->puser);
- lpar_submit ("system", (counter_t) data->psys);
- lpar_submit ("wait", (counter_t) data->pwait);
- lpar_submit ("idle", (counter_t) data->pidle);
-
- if (data->type.b.donate_enabled)
- {
- /* FYI: PURR == Processor Utilization of Resources Register
- * SPURR == Scaled PURR */
- lpar_submit ("idle_donated", (counter_t) data->idle_donated_purr);
- lpar_submit ("busy_donated", (counter_t) data->busy_donated_purr);
- lpar_submit ("idle_stolen", (counter_t) data->idle_stolen_purr);
- lpar_submit ("busy_stolen", (counter_t) data->busy_stolen_purr);
- }
-
- return (0);
-} /* int lpar_read_dedicated_partition */
+} /* void lpar_submit */
static int lpar_read (void)
{
perfstat_partition_total_t lparstats;
int status;
+ struct utsname name;
+ u_longlong_t ticks;
+ u_longlong_t user_ticks, syst_ticks, wait_ticks, idle_ticks;
+ u_longlong_t consumed_ticks;
+ double entitled_proc_capacity;
+
+ /* An LPAR has the same serial number as the physical system it is currently
+ running on. It is a convenient way of tracking LPARs as they are moved
+ from chassis to chassis through Live Partition Mobility (LPM). */
+ if (uname (&name) != 0)
+ {
+ ERROR ("lpar plugin: uname failed.");
+ return (-1);
+ }
+ sstrncpy (serial, name.machine, sizeof (serial));
/* Retrieve the current metrics. Returns the number of structures filled. */
status = perfstat_partition_total (/* name = */ NULL, /* (must be NULL) */
return (-1);
}
- if (lparstats.type.b.shared_enabled)
- lpar_read_shared_partition (&lparstats);
- else /* if (!shared_enabled) */
- lpar_read_dedicated_partition (&lparstats);
+ /* Number of ticks since we last run. */
+ ticks = lparstats.timebase_last - lparstats_old.timebase_last;
+ if (ticks == 0)
+ {
+ /* The stats have not been updated. Return now to avoid
+ * dividing by zero */
+ return (0);
+ }
+
+ /*
+ * On a shared partition, we're "entitled" to a certain amount of
+ * processing power, for example 250/100 of a physical CPU. Processing
+ * capacity not used by the partition may be assigned to a different
+ * partition by the hypervisor, so "idle" is hopefully a very small
+ * number.
+ *
+ * A dedicated partition may donate its CPUs to another partition and
+ * may steal ticks from somewhere else (another partition or maybe the
+ * shared pool, I don't know --octo).
+ */
+
+ /* entitled_proc_capacity is in 1/100th of a CPU */
+ entitled_proc_capacity = 0.01 * ((double) lparstats.entitled_proc_capacity);
+ lpar_submit ("entitled", entitled_proc_capacity);
+
+ /* The number of ticks actually spent in the various states */
+ user_ticks = lparstats.puser - lparstats_old.puser;
+ syst_ticks = lparstats.psys - lparstats_old.psys;
+ wait_ticks = lparstats.pwait - lparstats_old.pwait;
+ idle_ticks = lparstats.pidle - lparstats_old.pidle;
+ consumed_ticks = user_ticks + syst_ticks + wait_ticks + idle_ticks;
- if (pool_stats && !lparstats.type.b.pool_util_authority)
+ lpar_submit ("user", (double) user_ticks / (double) ticks);
+ lpar_submit ("system", (double) syst_ticks / (double) ticks);
+ lpar_submit ("wait", (double) wait_ticks / (double) ticks);
+ lpar_submit ("idle", (double) idle_ticks / (double) ticks);
+
+#if PERFSTAT_SUPPORTS_DONATION
+ if (donate_flag)
{
- WARNING ("lpar plugin: This partition does not have pool authority. "
- "Disabling CPU pool statistics collection.");
- pool_stats = 0;
+ /* donated => ticks given to another partition
+ * stolen => ticks received from another partition */
+ u_longlong_t idle_donated_ticks, busy_donated_ticks;
+ u_longlong_t idle_stolen_ticks, busy_stolen_ticks;
+
+ /* FYI: PURR == Processor Utilization of Resources Register
+ * SPURR == Scaled PURR */
+ idle_donated_ticks = lparstats.idle_donated_purr - lparstats_old.idle_donated_purr;
+ busy_donated_ticks = lparstats.busy_donated_purr - lparstats_old.busy_donated_purr;
+ idle_stolen_ticks = lparstats.idle_stolen_purr - lparstats_old.idle_stolen_purr;
+ busy_stolen_ticks = lparstats.busy_stolen_purr - lparstats_old.busy_stolen_purr;
+
+ lpar_submit ("idle_donated", (double) idle_donated_ticks / (double) ticks);
+ lpar_submit ("busy_donated", (double) busy_donated_ticks / (double) ticks);
+ lpar_submit ("idle_stolen", (double) idle_stolen_ticks / (double) ticks);
+ lpar_submit ("busy_stolen", (double) busy_stolen_ticks / (double) ticks);
+
+ /* Donated ticks will be accounted for as stolen ticks in other LPARs */
+ consumed_ticks += idle_stolen_ticks + busy_stolen_ticks;
}
+#endif
+
+ lpar_submit ("consumed", (double) consumed_ticks / (double) ticks);
if (pool_stats)
{
char typinst[DATA_MAX_NAME_LEN];
-
- /* Pool stats are in CPU x ns */
- ssnprintf (typinst, sizeof(typinst), "pool-%X-busy", lparstats.pool_id);
- lpar_submit (typinst, (double)lparstats.pool_busy_time / 1000000000.0);
-
- ssnprintf (typinst, sizeof(typinst), "pool-%X-total", lparstats.pool_id);
- lpar_submit (typinst, (double)lparstats.pool_max_time / 1000000000.0);
+ u_longlong_t pool_idle_cticks;
+ double pool_idle_cpus;
+ double pool_busy_cpus;
+
+ /* We're calculating "busy" from "idle" and the total number of
+ * CPUs, because the "busy" member didn't exist in early versions
+ * of libperfstat. It was added somewhere between AIX 5.3 ML5 and ML9. */
+ pool_idle_cticks = lparstats.pool_idle_time - lparstats_old.pool_idle_time;
+ pool_idle_cpus = CLOCKTICKS_TO_TICKS ((double) pool_idle_cticks) / (double) ticks;
+ pool_busy_cpus = ((double) lparstats.phys_cpus_pool) - pool_idle_cpus;
+ if (pool_busy_cpus < 0.0)
+ pool_busy_cpus = 0.0;
+
+ ssnprintf (typinst, sizeof (typinst), "pool-%X-busy", lparstats.pool_id);
+ lpar_submit (typinst, pool_busy_cpus);
+
+ ssnprintf (typinst, sizeof (typinst), "pool-%X-idle", lparstats.pool_id);
+ lpar_submit (typinst, pool_idle_cpus);
}
+ memcpy (&lparstats_old, &lparstats, sizeof (lparstats_old));
+
return (0);
} /* int lpar_read */
{
plugin_register_config ("lpar", lpar_config,
config_keys, config_keys_num);
+ plugin_register_init ("lpar", lpar_init);
plugin_register_read ("lpar", lpar_read);
} /* void module_register */
projects / collectd.git / blobdiff