pmseries, libpcp_web: add --gc mode to remove expired series data

man/man1/pmseries.1

@@ -18,7 +18,7 @@
 \f3pmseries\f1 \- display information about performance metric timeseries
 .SH SYNOPSIS
 \fBpmseries\fR
-[\fB\-adFiIlLmMnqsStvV?\fR]
+[\fB\-adFGiIlLmMnNqsStvV?\fR]
 [\fB\-c\fR \fIconfig\fR]
 [\f3\-D\f1 \f2debug\f1]
 [\fB\-g\fR \fIpattern\fR]
@@ -745,6 +745,36 @@ Connect to the key-value server at
 .IR host ,
 rather than the one the localhost.
 .TP
+\fB\-G\fR, \fB\-\-gc\fR
+Scan for expired time series and remove their ancillary data from the
+key-value cluster.
+A series is considered expired when its
+.B pcp:values:series:
+stream key no longer exists (i.e. the
+.B stream.expire
+TTL has elapsed without new data being ingested).
+The following keys are removed for each expired series:
+.B pcp:desc:series:,
+.B pcp:metric.name:series:,
+.B pcp:instances:series:,
+.B pcp:labelvalue:series:,
+and
+.B pcp:labelflags:series:.
+The series identifier is also removed from the forward-lookup sets
+.B pcp:series:metric.name:,
+.B pcp:series:inst.name:,
+.B pcp:series:label.<name>.value:,
+and
+.B pcp:series:context.name:.
+Orphaned
+.B pcp:inst:series:
+hashes are cleaned up when their associated instance-name set becomes empty.
+.TP
+\fB\-N\fR, \fB\-\-dryrun\fR
+Used with
+.BR \-\-gc :
+log which series would be removed without actually deleting anything.
+.TP
 \fB\-L\fR, \fB\-\-load\fR
 Load timeseries metadata and data into the key-value cluster.
 .TP

src/bashrc/pcp_completion.sh

@@ -137,7 +137,7 @@ _pcp_complete()
 	arg_regex="-[ahKcoFeDASTOstZiJ489N6XWwP0lEfqbyQBY]"
     ;;
     pmseries)
-	all_args="acdDFghiIlLmMnqpsStvVwZ"
+	all_args="acdDFgGhiIlLmMnNqpsStvVwZ"
 	arg_regex="-[cDghpwZ]"
     ;;
     pmstat)

src/include/pcp/pmwebapi.h

@@ -70,6 +70,7 @@ typedef enum pmSeriesFlags {
     PM_SERIES_FLAG_METADATA	= (1 << 0),	/* only load metric metadata */
     PM_SERIES_FLAG_ACTIVE	= (1 << 1),	/* continual source updates */
     PM_SERIES_FLAG_TEXT		= (1 << 2),	/* load metric & indom help */
+    PM_SERIES_FLAG_DRYRUN	= (1 << 3),	/* GC: log what would be removed, no writes */
     PM_SERIES_FLAG_ALL		= ((unsigned int)~PM_SERIES_FLAG_NONE)
 } pmSeriesFlags;
 
@@ -169,6 +170,7 @@ extern int pmSeriesValues(pmSeriesSettings *, pmSeriesTimeWindow *, int, sds *,
 extern int pmSeriesWindow(pmSeriesSettings *, sds, pmSeriesTimeWindow *, void *);
 extern int pmSeriesQuery(pmSeriesSettings *, sds, pmSeriesFlags, void *);
 extern int pmSeriesLoad(pmSeriesSettings *, sds, pmSeriesFlags, void *);
+extern int pmSeriesGC(pmSeriesSettings *, pmSeriesFlags, void *);
 
 /*
  * Timer list interface - global, thread-safe

src/libpcp_web/src/batons.h

@@ -25,6 +25,7 @@ typedef enum {
     MAGIC_NAMES,
     MAGIC_LABELMAP,
     MAGIC_SEARCH,
+    MAGIC_GC,
 
     MAGIC_COUNT
 } series_baton_magic;

src/libpcp_web/src/exports.in

@@ -294,3 +294,8 @@ PCP_WEB_1.23 {
     dictSetVal;
     keySlotsContextFree;
 } PCP_WEB_1.22;
+
+PCP_WEB_1.24 {
+  global:
+    pmSeriesGC;
+} PCP_WEB_1.23;