Skip to content

Commit

Permalink
pmdauwsgi: improve handling of transient uWSGI servers
Browse files Browse the repository at this point in the history 
Testing against a sample uWSGI server application from Sam,
this now seems to be resolved.  Minor changes to improve the
help text, handling of long-running global counters, and the
naming of worker instances.

Resolves Github issue #1964
  • Loading branch information
@natoscott
natoscott committed May 6, 2024
1 parent a856de2 commit 44f9017
Show file tree
Hide file tree
Showing 2 changed files with 26 additions and 18 deletions.
7 changes: 4 additions & 3 deletions qa/1992
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,11 +58,12 @@ pmdauwsgi_install()
echo
echo "=== $iam agent installation ==="
$sudo ./Install </dev/null >$tmp.out 2>&1
# Check metrics have appeared ... X metrics and Y values (or)
# Check metrics have appeared ... N warnings, X metrics and 0 values
cat $tmp.out >> $here/$seq.full
# Check uwsgi metrics have appeared ... X metrics and Y values (or)
# Check uwsgi metrics have appeared ... N warnings, X metrics and 0 values
_filter_pmda_install <$tmp.out \
| $PCP_AWK_PROG '
/Check uwsgi metrics have appeared/ { if ($NF > 10) {
/Check uwsgi metrics have appeared/ { if (NF > 12) {
if ($8 == "warnings,") $7 = $8 = ""
if ($9 >= 9) $9 = "X"
if ($12 == 0) $12 = "Y"
Expand Down
37 changes: 22 additions & 15 deletions src/pmdas/uwsgi/pmdauwsgi.python
View file
Original file line number Diff line number Diff line change
Expand Up @@ -121,6 +121,7 @@ class UwsgiPMDA(PMDA):

PMDA.__init__(self, name, domain)
self.timeout = DEFAULT_TIMEOUT
self.summary = UWSGISUMMARY()
self.host = DEFAULT_HOST
self.port = DEFAULT_PORT
self.get_failed = False
Expand All @@ -129,9 +130,7 @@ class UwsgiPMDA(PMDA):
self.connect_pmcd()

# Define the instance domains pointing to the dicts
#self.summary_indom = self.indom(0)
self.workers_indom = self.indom(0)
#self.add_indom(pmdaIndom(self.summary_indom, self.summary))
self.add_indom(pmdaIndom(self.workers_indom, self.workers))

# define the Summary (roll up, at-a-glance numbers)
Expand All @@ -147,14 +146,14 @@ class UwsgiPMDA(PMDA):
# .total_pause_worker_count

self.add_metric(name + '.summary.total_workers', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 0), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Total number of uwsgi workers")
self.add_metric(name + '.summary.avg_response_time_msec', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 1), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Average response time across all workers")
self.add_metric(name + '.summary.total_requests_served', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 2), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Total requests served across all workers")
self.add_metric(name + '.summary.total_workers_accepting_requests', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 3), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Total number of workers accepting requests")
self.add_metric(name + '.summary.total_exceptions', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 4), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Total exceptions across all workers")
self.add_metric(name + '.summary.total_harakiri_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 5), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Total harakiri count across all workers")
self.add_metric(name + '.summary.total_busy_worker_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 6), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Total number of busy workers")
self.add_metric(name + '.summary.total_idle_worker_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 7), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Total number of idle workers")
self.add_metric(name + '.summary.total_pause_worker_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 8), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Total number of pause workers")
self.add_metric(name + '.summary.avg_response_time_msec', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 1), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Average response time across all current workers")
self.add_metric(name + '.summary.total_requests_served', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 2), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Total requests served by all workers since starting")
self.add_metric(name + '.summary.total_workers_accepting_requests', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 3), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Number of current workers accepting requests")
self.add_metric(name + '.summary.total_exceptions', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 4), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Total exceptions across all workers since starting")
self.add_metric(name + '.summary.total_harakiri_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 5), PM_TYPE_U64, PM_INDOM_NULL, PM_SEM_COUNTER, pmUnits()), "Total harakiri count across all workers since starting")
self.add_metric(name + '.summary.total_busy_worker_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 6), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Number of busy workers currently")
self.add_metric(name + '.summary.total_idle_worker_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 7), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Number of idle workers currently")
self.add_metric(name + '.summary.total_pause_worker_count', pmdaMetric(PMDA.pmid(UwsgiPMDA.UWSGI_SUMMARY, 8), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT, pmUnits()), "Number of pause workers currently")

# Add the uwsgi worker metrics (metrics per worker)
# uwsgi.worker
Expand All @@ -179,16 +178,21 @@ class UwsgiPMDA(PMDA):
self.set_user(PCP.pmGetConfig('PCP_USER'))

def refresh_all(self):
stats = {}
stats['workers'] = []

try:
stats = requests.get(self.url, timeout=self.timeout).json()
self.get_failed = False
#self.log("refresh_all stats:", str(stats))
self.get_failed = False
except Exception:
self.get_failed = True
return

avg_response_time_msec = 0
summary = UWSGISUMMARY()
# pass any persistent long-running counters into the constructor
summary = UWSGISUMMARY(req=self.summary.total_requests_served,
harakiri=self.summary.total_harakiri_count,
exception=self.summary.total_exceptions)
for w in stats['workers']:
worker_id = w['id']
worker = WORKER()
Expand All @@ -212,8 +216,11 @@ class UwsgiPMDA(PMDA):
elif worker.status == 'pause':
summary.total_pause_worker_count += 1
avg_response_time_msec += worker.avg_rt
self.workers[str(worker_id)] = worker
summary.avg_response_time_msec = avg_response_time_msec/summary.total_workers
self.workers['worker-'+str(worker_id)] = worker

summary.avg_response_time_msec = avg_response_time_msec
if summary.total_workers > 1: # divide-by-zero guard
summary.avg_response_time_msec /= summary.total_workers
self.summary = summary

def read_config(self):
Expand Down

0 comments on commit 44f9017

Please sign in to comment.