From b9fc63c362c5176298ce0a0a903b7ecbfb20a770 Mon Sep 17 00:00:00 2001
From: Vihas Makwana <121151420+VihasMakwana@users.noreply.github.com>
Date: Tue, 24 Dec 2024 16:26:10 +0530
Subject: [PATCH] [process] - Add a boolean to detect partial matches (#199)

This PR follows up on
https://github.com/elastic/elastic-agent-system-metrics/pull/195 and
adds a new boolean to indicate if a given process state is partial.
This is useful while returning errors to the caller.
---
 metric/system/process/process.go       | 11 +++++++++--
 metric/system/process/process_types.go |  4 +++-
 2 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/metric/system/process/process.go b/metric/system/process/process.go
index 7045af80d..698e98236 100644
--- a/metric/system/process/process.go
+++ b/metric/system/process/process.go
@@ -140,11 +140,11 @@ func (procStats *Stats) Get() ([]mapstr.M, []mapstr.M, error) {
 		procs = append(procs, proc)
 		rootEvents = append(rootEvents, rootMap)
 	}
-	if len(failedPIDs) > 0 {
+	if wrappedErr != nil && len(failedPIDs) > 0 {
 		procStats.logger.Debugf("error fetching process metrics: %v", wrappedErr)
 		return procs, rootEvents, NonFatalErr{Err: fmt.Errorf(errFetchingPIDs, len(failedPIDs))}
 	}
-	return procs, rootEvents, nil
+	return procs, rootEvents, toNonFatal(wrappedErr)
 }
 
 // GetOne fetches process data for a given PID if its name matches the regexes provided from the host.
@@ -224,6 +224,10 @@ func (procStats *Stats) pidIter(pid int, procMap ProcsMap, proclist []ProcState)
 		procStats.logger.Debugf("Process name does not match the provided regex; PID=%d; name=%s", pid, status.Name)
 		return procMap, proclist, nonFatalErr
 	}
+	// there was some non-fatal error and given state is partial
+	if nonFatalErr != nil {
+		status.Partial = true
+	}
 	procMap[pid] = status
 	proclist = append(proclist, status)
 
@@ -422,12 +426,15 @@ func (procStats *Stats) isWhitelistedEnvVar(varName string) bool {
 }
 
 func extractFailedPIDs(procMap ProcsMap) []int {
+	// calculate the total amount of partial/failed PIDs
 	list := make([]int, 0)
 	for pid, state := range procMap {
 		if state.Failed {
 			list = append(list, pid)
 			// delete the failed state so we don't return the state to caller
 			delete(procMap, pid)
+		} else if state.Partial {
+			list = append(list, pid)
 		}
 	}
 	return list
diff --git a/metric/system/process/process_types.go b/metric/system/process/process_types.go
index 16d7e2426..e66cfb538 100644
--- a/metric/system/process/process_types.go
+++ b/metric/system/process/process_types.go
@@ -57,8 +57,10 @@ type ProcState struct {
 	// meta
 	SampleTime time.Time `struct:"-,omitempty"`
 
-	// boolean to indicate that given PID has failed due to some error.
+	// boolean to indicate that given PID has completeley failed due to some error.
 	Failed bool `struct:"-,omitempty"`
+	// boolean to indicate that given state is partially filled.
+	Partial bool `struct:"-,omitempty"`
 }
 
 // ProcCPUInfo is the main struct for CPU metrics