From 8860ada0a6afd6adb2b26d5b0ab161bfe66c6019 Mon Sep 17 00:00:00 2001
From: Tom Clegg <tom@curii.com>
Date: Tue, 25 Oct 2022 14:00:38 -0400
Subject: [PATCH] 19563: Report process size in bytes, not pages.

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curii.com>
---
 lib/crunchstat/crunchstat.go      | 32 +++++++++++++++++++++++++++++--
 lib/crunchstat/crunchstat_test.go | 10 +++++++++-
 2 files changed, 39 insertions(+), 3 deletions(-)

diff --git a/lib/crunchstat/crunchstat.go b/lib/crunchstat/crunchstat.go
index 443d2202ce..3a473cab87 100644
--- a/lib/crunchstat/crunchstat.go
+++ b/lib/crunchstat/crunchstat.go
@@ -14,6 +14,7 @@ import (
 	"io"
 	"io/ioutil"
 	"os"
+	"regexp"
 	"sort"
 	"strconv"
 	"strings"
@@ -52,6 +53,7 @@ type Reporter struct {
 		Printf(fmt string, args ...interface{})
 	}
 
+	kernelPageSize      int64
 	reportedStatFile    map[string]string
 	lastNetSample       map[string]ioSample
 	lastDiskIOSample    map[string]ioSample
@@ -274,6 +276,32 @@ func (r *Reporter) doMemoryStats() {
 	}
 	r.Logger.Printf("mem%s\n", outstat.String())
 
+	if r.kernelPageSize == 0 {
+		// assign "don't try again" value in case we give up
+		// and return without assigning the real value
+		r.kernelPageSize = -1
+		buf, err := os.ReadFile("/proc/self/smaps")
+		if err != nil {
+			r.Logger.Printf("error reading /proc/self/smaps: %s", err)
+			return
+		}
+		m := regexp.MustCompile(`\nKernelPageSize:\s*(\d+) kB\n`).FindSubmatch(buf)
+		if len(m) != 2 {
+			r.Logger.Printf("error parsing /proc/self/smaps: KernelPageSize not found")
+			return
+		}
+		size, err := strconv.ParseInt(string(m[1]), 10, 64)
+		if err != nil {
+			r.Logger.Printf("error parsing /proc/self/smaps: KernelPageSize %q: %s", m[1], err)
+			return
+		}
+		r.kernelPageSize = size * 1024
+	} else if r.kernelPageSize < 0 {
+		// already failed to determine page size, don't keep
+		// trying/logging
+		return
+	}
+
 	r.reportPIDsMu.Lock()
 	defer r.reportPIDsMu.Unlock()
 	procnames := make([]string, 0, len(r.reportPIDs))
@@ -303,11 +331,11 @@ func (r *Reporter) doMemoryStats() {
 		// rss is the 24th field in .../stat, and fields[0]
 		// here is the last char ')' of the 2nd field, so
 		// rss is fields[22]
-		rss, err := strconv.Atoi(string(fields[22]))
+		rss, err := strconv.ParseInt(string(fields[22]), 10, 64)
 		if err != nil {
 			continue
 		}
-		procmem += fmt.Sprintf(" %d %s", rss, procname)
+		procmem += fmt.Sprintf(" %d %s", rss*r.kernelPageSize, procname)
 	}
 	if procmem != "" {
 		r.Logger.Printf("procmem%s\n", procmem)
diff --git a/lib/crunchstat/crunchstat_test.go b/lib/crunchstat/crunchstat_test.go
index 922aa369ba..5e8e93de6c 100644
--- a/lib/crunchstat/crunchstat_test.go
+++ b/lib/crunchstat/crunchstat_test.go
@@ -9,6 +9,7 @@ import (
 	"log"
 	"os"
 	"regexp"
+	"strconv"
 	"testing"
 	"time"
 
@@ -69,7 +70,14 @@ func (s *suite) TestReportPIDs(c *C) {
 			c.Error("timed out")
 			break
 		}
-		if regexp.MustCompile(`(?ms).*procmem \d+ init \d+ test_process.*`).MatchString(logbuf.String()) {
+		if m := regexp.MustCompile(`(?ms).*procmem \d+ init (\d+) test_process.*`).FindSubmatch(logbuf.Bytes()); len(m) > 0 {
+			size, err := strconv.ParseInt(string(m[1]), 10, 64)
+			c.Check(err, IsNil)
+			// Expect >1 MiB and <100 MiB -- otherwise we
+			// are probably misinterpreting /proc/N/stat
+			// or multiplying by the wrong page size.
+			c.Check(size > 1000000, Equals, true)
+			c.Check(size < 100000000, Equals, true)
 			break
 		}
 	}
-- 
2.30.2