fix: move live Statfs to RemainingResources; use *uint64 for min-free config

- TotalResources reverts to static startup value to keep total-remaining
  arithmetic consistent for BBS metrics
- RemainingResources caps reported DiskMB at live partition free space
  (syscall.Statfs), so schedulers see reduced capacity when disk fills
- MinCachePartitionFreeBytes changed to *uint64: nil means use the 5 GB
  default, 0 explicitly disables the check (previously impossible)

ai-assisted=yes
TNZ-111552

Author: kart2bc

jobs/rep/spec

@@ -147,6 +147,8 @@ properties:
   diego.executor.max_cache_size_in_bytes:
     description: "maximum size of the cache in bytes - this should leave a healthy overhead for temporary items, etc."
     default: 10000000000
+  diego.executor.min_cache_partition_free_bytes:
+    description: "Minimum free space in bytes that must remain on the disk where the download cache lives. When actual disk free space falls below this threshold, the cache deletes old downloaded files to recover space. Uses live partition statistics rather than in-memory accounting. Defaults to 5368709120 (5 GB) if unset."
   diego.executor.use_schedulable_disk_size:
     description: "Use total space available to containers reported by Garden. If false the total size of image plugin store minus max_cache_size_in_bytes is used."
     default: false

jobs/rep/templates/rep.json.erb

@@ -125,6 +125,10 @@
     raise "additional_memory_allocation_mb cannot be negative"
   end
 
+  if_p("diego.executor.min_cache_partition_free_bytes") do |value|
+    config[:min_cache_partition_free_bytes] = value
+  end
+
   if_p("diego.executor.volman.driver_paths") do |value|
     config[:volman_driver_paths] = value
   end

jobs/rep_windows/spec

@@ -129,6 +129,8 @@ properties:
   diego.executor.max_cache_size_in_bytes:
     description: "maximum size of the cache in bytes - this should leave a healthy overhead for temporary items, etc."
     default: 10000000000
+  diego.executor.min_cache_partition_free_bytes:
+    description: "Minimum free space in bytes that must remain on the disk where the download cache lives. When actual disk free space falls below this threshold, the cache deletes old downloaded files to recover space. Currently does not function on Windows cells. Defaults to 5368709120 (5 GB) if unset."
   diego.executor.use_schedulable_disk_size:
     description: "Use total space available to containers reported by Garden. If false the total size of image plugin store minus max_cache_size_in_bytes is used."
     default: false

jobs/rep_windows/templates/rep.json.erb

@@ -122,6 +122,10 @@
     raise "additional_memory_allocation_mb cannot be negative"
   end
 
+  if_p("diego.executor.min_cache_partition_free_bytes") do |value|
+    config[:min_cache_partition_free_bytes] = value
+  end
+
   if_p("diego.executor.volman.driver_paths") do |value|
     config[:volman_driver_paths] = value
   end

src/code.cloudfoundry.org/cacheddownloader/file_cache.go

@@ -4,11 +4,9 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"math"
 	"os"
 	"path/filepath"
 	"sync"
-	"syscall"
 	"time"
 
 	"code.cloudfoundry.org/archiver/compressor"
@@ -34,14 +32,6 @@ type FileCache struct {
 	Seq                       uint64
 }
 
-func defaultFreeSpaceOnPartition(path string) int64 {
-	var stat syscall.Statfs_t
-	if err := syscall.Statfs(path, &stat); err != nil {
-		return math.MaxInt64
-	}
-	return int64(stat.Bavail) * int64(stat.Bsize)
-}
-
 type FileCacheEntry struct {
 	Size                  int64
 	Access                time.Time
@@ -201,6 +191,11 @@ func (e *FileCacheEntry) expandedDirectory() (string, error) {
 			err = os.RemoveAll(e.FilePath)
 			if err != nil {
 				fmt.Fprintln(os.Stderr, "Unable to delete the cached file", err)
+			} else {
+				// GetDirectory doubled Size before calling here; tar is now gone so halve it
+				// back to reflect only the .tar.d on disk. Mirrors the symmetric halving in
+				// decrementFileInUseCount() which fires when the tar is deleted later.
+				e.Size = e.Size / 2
 			}
 		}
 	}

src/code.cloudfoundry.org/cacheddownloader/file_cache_unix.go

@@ -0,0 +1,16 @@
+//go:build !windows
+
+package cacheddownloader
+
+import (
+	"math"
+	"syscall"
+)
+
+func defaultFreeSpaceOnPartition(path string) int64 {
+	var stat syscall.Statfs_t
+	if err := syscall.Statfs(path, &stat); err != nil {
+		return math.MaxInt64
+	}
+	return int64(stat.Bavail) * int64(stat.Bsize)
+}

src/code.cloudfoundry.org/cacheddownloader/file_cache_windows.go

@@ -0,0 +1,9 @@
+//go:build windows
+
+package cacheddownloader
+
+import "math"
+
+func defaultFreeSpaceOnPartition(path string) int64 {
+	return math.MaxInt64
+}

src/code.cloudfoundry.org/executor/depot/depot.go

@@ -219,17 +219,29 @@ func (c *client) DeleteContainer(logger lager.Logger, traceID string, guid strin
 
 func (c *client) RemainingResources(logger lager.Logger) (executor.ExecutorResources, error) {
 	logger = logger.Session("remaining-resources")
-	return c.containerStore.RemainingResources(logger), nil
+	remaining := c.containerStore.RemainingResources(logger)
+	if c.diskPath != "" {
+		if liveMB, ok := liveDiskMB(c.diskPath); ok && liveMB < remaining.DiskMB {
+			remaining.DiskMB = liveMB
+		}
+	}
+	return remaining, nil
 }
 
 func (c *client) Ping(logger lager.Logger) error {
 	return c.gardenClient.Ping()
 }
 
 func (c *client) TotalResources(logger lager.Logger) (executor.ExecutorResources, error) {
+	diskMB := c.totalCapacity.DiskMB
+	if c.diskPath != "" {
+		if liveMB, ok := liveDiskMB(c.diskPath); ok {
+			diskMB = liveMB
+		}
+	}
 	return executor.ExecutorResources{
 		MemoryMB:   c.totalCapacity.MemoryMB,
-		DiskMB:     getDiskMB(c.diskPath, c.totalCapacity.DiskMB),
+		DiskMB:     diskMB,
 		Containers: c.totalCapacity.Containers,
 	}, nil
 }

src/code.cloudfoundry.org/executor/depot/depot_test.go

@@ -698,45 +698,70 @@ var _ = Describe("Depot", func() {
 	})
 
 	Describe("RemainingResources", func() {
-		var resources executor.ExecutorResources
+		var remainingResources executor.ExecutorResources
 
 		BeforeEach(func() {
-			resources = executor.NewExecutorResources(1024, 1024, 3)
-			containerStore.RemainingResourcesReturns(resources)
+			remainingResources = executor.NewExecutorResources(1024, 1024, 3)
+			containerStore.RemainingResourcesReturns(remainingResources)
 		})
 
 		It("should reduce resources used by allocated and running containers", func() {
 			actualResources, err := depotClient.RemainingResources(logger)
 			Expect(err).NotTo(HaveOccurred())
-			Expect(actualResources).To(Equal(resources))
+			Expect(actualResources).To(Equal(remainingResources))
+		})
+
+		Context("when disk path is configured and partition has less space than the store reports", func() {
+			var tmpDir string
+
+			BeforeEach(func() {
+				var err error
+				tmpDir, err = os.MkdirTemp("", "depot-remaining-disk-test")
+				Expect(err).NotTo(HaveOccurred())
+				diskPath = tmpDir
+				containerStore.RemainingResourcesReturns(executor.NewExecutorResources(1024, math.MaxInt32, 3))
+			})
+
+			It("caps remaining disk at the live partition free space from syscall.Statfs", func() {
+				result, err := depotClient.RemainingResources(logger)
+				Expect(err).NotTo(HaveOccurred())
+				Expect(result.DiskMB).NotTo(Equal(math.MaxInt32))
+				Expect(result.DiskMB).To(BeNumerically(">", 0))
+			})
 		})
 	})
 
 	Describe("TotalResources", func() {
-		Context("when asked for total resources", func() {
-			Context("when disk path is not provided", func() {
-				It("should return the resources it was configured with", func() {
-					Expect(depotClient.TotalResources(logger)).To(Equal(resources))
-				})
+		Context("when no disk path is configured", func() {
+			It("returns the static startup resources", func() {
+				result, err := depotClient.TotalResources(logger)
+				Expect(err).NotTo(HaveOccurred())
+				Expect(result).To(Equal(resources))
 			})
+		})
 
-			Context("when disk path is provided", func() {
-				BeforeEach(func() {
-					diskPath = os.TempDir()
-				})
+		Context("when disk path is configured", func() {
+			var tmpDir string
 
-				It("should return the resources it was configured with, with live disk capacity", func() {
-					totalResources, err := depotClient.TotalResources(logger)
-					Expect(err).NotTo(HaveOccurred())
+			BeforeEach(func() {
+				var err error
+				tmpDir, err = os.MkdirTemp("", "depot-total-disk-test")
+				Expect(err).NotTo(HaveOccurred())
+				diskPath = tmpDir
+				resources.DiskMB = math.MaxInt32
+			})
 
-					Expect(totalResources.MemoryMB).To(Equal(resources.MemoryMB))
-					Expect(totalResources.Containers).To(Equal(resources.Containers))
+			AfterEach(func() {
+				os.RemoveAll(tmpDir)
+			})
 
-					// disk capacity should be updated
-					Expect(totalResources.DiskMB).NotTo(Equal(resources.DiskMB))
-					Expect(totalResources.DiskMB).To(BeNumerically(">", 0))
-					Expect(totalResources.DiskMB).To(BeNumerically("<=", math.MaxInt32))
-				})
+			It("returns the resources it was configured with, with live disk capacity", func() {
+				result, err := depotClient.TotalResources(logger)
+				Expect(err).NotTo(HaveOccurred())
+				Expect(result.MemoryMB).To(Equal(resources.MemoryMB))
+				Expect(result.Containers).To(Equal(resources.Containers))
+				Expect(result.DiskMB).NotTo(Equal(math.MaxInt32))
+				Expect(result.DiskMB).To(BeNumerically(">", 0))
 			})
 		})
 	})

src/code.cloudfoundry.org/executor/depot/depot_unix.go

@@ -1,17 +1,13 @@
 //go:build !windows
-// +build !windows
 
 package depot
 
 import "syscall"
 
-func getDiskMB(diskPath string, fallback int) int {
-	if diskPath == "" {
-		return fallback
-	}
+func liveDiskMB(path string) (int, bool) {
 	var stat syscall.Statfs_t
-	if err := syscall.Statfs(diskPath, &stat); err == nil {
-		return int(int64(stat.Bavail) * int64(stat.Bsize) / (1024 * 1024))
+	if err := syscall.Statfs(path, &stat); err != nil {
+		return 0, false
 	}
-	return fallback
+	return int(int64(stat.Bavail) * int64(stat.Bsize) / (1024 * 1024)), true
 }