if we are low on disk space, only run GC if it will remove any data (#603)

* if we are low on disk space, only run GC if it will remove any data
* make sure we don't treat bail of GC due to disk space as success causing 0 wait between GC passes

1 files changed, 157 insertions, 89 deletions

diff --git a/src/zenstore/gc.cpp b/src/zenstore/gc.cpp
index 3fc795f9e..8c3d802c3 100644
--- a/src/zenstore/gc.cpp
+++ b/src/zenstore/gc.cpp
@@ -58,6 +58,8 @@ namespace fs = std::filesystem;
 //////////////////////////////////////////////////////////////////////////
 
 namespace {
+	static std::filesystem::path GcDiskReserveFileName = "reserve.gc";
+
 	std::error_code CreateGCReserve(const std::filesystem::path& Path, uint64_t Size)
 	{
 		if (Size == 0)
@@ -1678,7 +1680,7 @@ DiskUsageWindow::GetDiskDeltas(GcClock::Tick StartTick, GcClock::Tick EndTick, G
 }
 
 GcClock::Tick
-DiskUsageWindow::FindTimepointThatRemoves(uint64_t Amount, GcClock::Tick EndTick) const
+DiskUsageWindow::FindTimepointThatRemoves(uint64_t Amount, GcClock::Tick EndTick, uint64_t& OutFoundAmount) const
 {
 	ZEN_ASSERT(Amount > 0);
 	uint64_t RemainingToFind = Amount;
@@ -1688,17 +1690,29 @@ DiskUsageWindow::FindTimepointThatRemoves(uint64_t Amount, GcClock::Tick EndTick
 		const DiskUsageEntry& Entry = m_LogWindow[Offset];
 		if (Entry.SampleTime >= EndTick)
 		{
+			OutFoundAmount = Amount - RemainingToFind;
+			if (OutFoundAmount > 0)
+			{
+				return m_LogWindow[Offset - 1].SampleTime + 1;
+			}
 			return EndTick;
 		}
 		const DiskUsageEntry& PreviousEntry = m_LogWindow[Offset - 1];
 		uint64_t			  Delta			= Entry.DiskUsage > PreviousEntry.DiskUsage ? Entry.DiskUsage - PreviousEntry.DiskUsage : 0;
 		if (Delta >= RemainingToFind)
 		{
+			OutFoundAmount = (Amount - RemainingToFind) + Delta;
 			return m_LogWindow[Offset].SampleTime + 1;
 		}
 		RemainingToFind -= Delta;
 		Offset++;
 	}
+	OutFoundAmount = Amount - RemainingToFind;
+	if (OutFoundAmount > 0)
+	{
+		// Remove what we can if we found anything
+		return m_LogWindow[Offset - 1].SampleTime + 1;
+	}
 	return EndTick;
 }
 
@@ -1737,11 +1751,11 @@ GcScheduler::Initialize(const GcSchedulerConfig& Config)
 
 	CreateDirectories(Config.RootDirectory);
 
-	std::error_code Ec = CreateGCReserve(m_Config.RootDirectory / "reserve.gc", m_Config.DiskReserveSize);
+	std::error_code Ec = CreateGCReserve(m_Config.RootDirectory / GcDiskReserveFileName, m_Config.DiskReserveSize);
 	if (Ec)
 	{
 		ZEN_WARN("unable to create GC reserve at '{}' with size {}, reason '{}'",
-				 m_Config.RootDirectory / "reserve.gc",
+				 m_Config.RootDirectory / GcDiskReserveFileName,
 				 NiceBytes(m_Config.DiskReserveSize),
 				 Ec.message());
 	}
@@ -2029,7 +2043,7 @@ GcScheduler::GetState() const
 	if (Result.Config.DiskReserveSize != 0)
 	{
 		Ec.clear();
-		Result.HasDiskReserve = IsFile(Result.Config.RootDirectory / "reserve.gc", Ec) && !Ec;
+		Result.HasDiskReserve = IsFile(Result.Config.RootDirectory / GcDiskReserveFileName, Ec) && !Ec;
 	}
 
 	if (Result.Status != GcSchedulerStatus::kRunning)
@@ -2330,20 +2344,44 @@ GcScheduler::SchedulerThread()
 					const uint64_t GcDiskSpaceRemoveGoal = Max(MaximumDiskUseGcSpaceGoal, MinimumFreeDiskGcSpaceGoal);
 
 					std::unique_lock Lock(m_GcMutex);
+					uint64_t		 FoundAmount = 0;
+					GcClock::Tick	 NowTick	 = Now.time_since_epoch().count();
 					GcClock::Tick	 AgeTick =
-						m_DiskUsageWindow.FindTimepointThatRemoves(GcDiskSpaceRemoveGoal, Now.time_since_epoch().count());
-					GcClock::TimePoint SizeBasedExpireTime = GcClock::TimePointFromTick(AgeTick);
-					if (SizeBasedExpireTime > CacheExpireTime)
-					{
-						CacheExpireTime = SizeBasedExpireTime;
-					}
-					if (SizeBasedExpireTime > ProjectStoreExpireTime)
+						m_DiskUsageWindow.FindTimepointThatRemoves(GcDiskSpaceRemoveGoal, Now.time_since_epoch().count(), FoundAmount);
+					if (AgeTick < NowTick)
 					{
-						ProjectStoreExpireTime = SizeBasedExpireTime;
+						ZEN_ASSERT(FoundAmount > 0);
+						GcClock::TimePoint SizeBasedExpireTime = GcClock::TimePointFromTick(AgeTick);
+						if (SizeBasedExpireTime > CacheExpireTime)
+						{
+							CacheExpireTime = SizeBasedExpireTime;
+						}
+						if (SizeBasedExpireTime > ProjectStoreExpireTime)
+						{
+							ProjectStoreExpireTime = SizeBasedExpireTime;
+						}
+						if (SizeBasedExpireTime > BuildStoreExpireTime)
+						{
+							BuildStoreExpireTime = SizeBasedExpireTime;
+						}
 					}
-					if (SizeBasedExpireTime > BuildStoreExpireTime)
+					else
 					{
-						BuildStoreExpireTime = SizeBasedExpireTime;
+						if (HighDiskSpaceUsageGCTriggered)
+						{
+							ZEN_WARN("Used disk space {} is above {} but no data available to free",
+									 NiceBytes(TotalSize.DiskSize),
+									 NiceBytes(DiskSizeSoftLimit));
+							HighDiskSpaceUsageGCTriggered = false;
+						}
+						if (LowDiskSpaceGCTriggered)
+						{
+							ZEN_WARN("Free disk space {} is below {} but no data available to free",
+									 NiceBytes(Space.Free),
+									 NiceBytes(MinimumFreeDiskSpaceToAllowWrites));
+							LowDiskSpaceGCTriggered = false;
+							m_AreDiskWritesBlocked.store(true);
+						}
 					}
 				}
 
@@ -2476,34 +2514,44 @@ GcScheduler::SchedulerThread()
 				m_AttachmentPassIndex = NextAttachmentPassIndex;
 			}
 
-			bool GcSuccess = CollectGarbage(CacheExpireTime,
-											ProjectStoreExpireTime,
-											BuildStoreExpireTime,
-											DoDelete,
-											CollectSmallObjects,
-											SkipCid,
-											UseGCVersion,
-											CompactBlockUsageThresholdPercent,
-											Verbose,
-											SingleThreaded,
-											AttachmentRangeMin,
-											AttachmentRangeMax,
-											StoreCacheAttachmentMetaData,
-											StoreProjectAttachmentMetaData,
-											EnableValidation,
-											SilenceErrors);
-			if (!GcSuccess)
+			if (PrepareDiskReserve())
 			{
-				SilenceErrors = true;
-				ZEN_INFO("gc going into error state - gc errors will be demoted to warnings until we recover");
+				bool GcSuccess = CollectGarbage(CacheExpireTime,
+												ProjectStoreExpireTime,
+												BuildStoreExpireTime,
+												DoDelete,
+												CollectSmallObjects,
+												SkipCid,
+												UseGCVersion,
+												CompactBlockUsageThresholdPercent,
+												Verbose,
+												SingleThreaded,
+												AttachmentRangeMin,
+												AttachmentRangeMax,
+												StoreCacheAttachmentMetaData,
+												StoreProjectAttachmentMetaData,
+												EnableValidation,
+												SilenceErrors);
+				if (!GcSuccess)
+				{
+					SilenceErrors = true;
+					ZEN_INFO("gc going into error state - gc errors will be demoted to warnings until we recover");
+				}
+				else if (SilenceErrors)
+				{
+					SilenceErrors = false;
+					ZEN_INFO("gc recovered from error state - gc errors will be will be reported as errors again");
+				}
+
+				WaitTime = std::chrono::seconds(0);
 			}
-			else if (SilenceErrors)
+			else
 			{
-				SilenceErrors = false;
-				ZEN_INFO("gc recovered from error state - gc errors will be will be reported as errors again");
+				ZEN_WARN(
+					"Disk space is very low and we have no GC reserve, skipping GC as this requires at least some space to write to '{}'",
+					m_Config.RootDirectory);
+				WaitTime = m_Config.MonitorInterval;
 			}
-
-			WaitTime = std::chrono::seconds(0);
 		}
 		catch (const std::system_error& SystemError)
 		{
@@ -2582,6 +2630,63 @@ GcScheduler::ScrubStorage(bool DoDelete, bool SkipCid, std::chrono::seconds Time
 }
 
 bool
+GcScheduler::ReclaimDiskReserve()
+{
+	const std::filesystem::path DiskReservePath = m_Config.RootDirectory / GcDiskReserveFileName;
+	std::error_code				Ec				= CreateGCReserve(DiskReservePath, m_Config.DiskReserveSize);
+	if (Ec)
+	{
+		ZEN_WARN("unable to create GC reserve at '{}' with size {}, reason: '{}'",
+				 DiskReservePath,
+				 NiceBytes(m_Config.DiskReserveSize),
+				 Ec.message());
+		return false;
+	}
+	return true;
+}
+
+bool
+GcScheduler::PrepareDiskReserve()
+{
+	try
+	{
+		static bool ForceFail = false;
+		if (ForceFail)
+		{
+			return false;
+		}
+		(void)ReclaimDiskReserve();
+		CheckDiskSpace();
+
+		if (m_AreDiskWritesBlocked.load())
+		{
+			// We are low on disk, check if we can release our extra storage reserve, if we can't bail from doing GC
+			uint64_t					ReleasedSpace	= 0;
+			const std::filesystem::path DiskReservePath = m_Config.RootDirectory / GcDiskReserveFileName;
+			if (IsFile(DiskReservePath))
+			{
+				uint64_t GcReserveFileSize = FileSizeFromPath(DiskReservePath);
+				if (RemoveFile(DiskReservePath))
+				{
+					ReleasedSpace = GcReserveFileSize;
+				}
+			}
+
+			if (ReleasedSpace == 0)
+			{
+				return false;
+			}
+		}
+		return true;
+	}
+	catch (const std::exception& Ex)
+	{
+		ZEN_WARN("Failed to prepare for GC, reason: {}", Ex.what());
+		return false;
+	}
+}
+
+bool
 GcScheduler::CollectGarbage(const GcClock::TimePoint& CacheExpireTime,
 							const GcClock::TimePoint& ProjectStoreExpireTime,
 							const GcClock::TimePoint& BuildStoreExpireTime,
@@ -2605,53 +2710,8 @@ GcScheduler::CollectGarbage(const GcClock::TimePoint& CacheExpireTime,
 
 	try
 	{
-		const std::filesystem::path DiskReservePath = m_Config.RootDirectory / "reserve.gc";
-
-		auto ReclaimDiskReserve = [&]() {
-			std::error_code Ec = CreateGCReserve(DiskReservePath, m_Config.DiskReserveSize);
-			if (Ec)
-			{
-				ZEN_WARN("unable to create GC reserve at '{}' with size {}, reason: '{}'",
-						 m_Config.RootDirectory / "reserve.gc",
-						 NiceBytes(m_Config.DiskReserveSize),
-						 Ec.message());
-			}
-		};
-
-		ReclaimDiskReserve();
 		const auto _ = MakeGuard([&] { ReclaimDiskReserve(); });
 
-		CheckDiskSpace();
-
-		if (m_AreDiskWritesBlocked.load())
-		{
-			// We are low on disk, check if we can release our extra storage reserve, if we can't bail from doing GC
-			auto ClaimDiskReserve = [&]() -> uint64_t {
-				if (!IsFile(DiskReservePath))
-				{
-					return 0;
-				}
-				uint64_t ReclaimedSize = FileSizeFromPath(DiskReservePath);
-				if (RemoveFile(DiskReservePath))
-				{
-					return ReclaimedSize;
-				}
-				return 0;
-			};
-
-			uint64_t ReleasedSpace = ClaimDiskReserve();
-			if (ReleasedSpace == 0)
-			{
-				ZEN_WARN(
-					"Disk space is very low and we have no GC reserve, skipping GC as this requires at least some space to write to '{}'",
-					m_Config.RootDirectory);
-				m_LastGcTime			= GcClock::Now();
-				m_LastLightweightGcTime = m_LastGcTime;
-				return true;  // Treat this as a success as we don't want to silence any errors that happens when we are no longer low on
-							  // disk space
-			}
-		}
-
 		{
 			Stopwatch  Timer;
 			const auto __ = MakeGuard([&] { ZEN_INFO("garbage collection DONE in {}", NiceTimeSpanMs(Timer.GetElapsedTimeMs())); });
@@ -2675,7 +2735,7 @@ GcScheduler::CollectGarbage(const GcClock::TimePoint& CacheExpireTime,
 													 .Verbose							= Verbose,
 													 .SingleThread						= SingleThreaded,
 													 .CompactBlockUsageThresholdPercent = CompactBlockUsageThresholdPercent,
-													 .DiskReservePath					= DiskReservePath,
+													 .DiskReservePath					= m_Config.RootDirectory / GcDiskReserveFileName,
 													 .AttachmentRangeMin				= AttachmentRangeMin,
 													 .AttachmentRangeMax				= AttachmentRangeMax,
 													 .StoreCacheAttachmentMetaData		= StoreCacheAttachmentMetaData,
@@ -3095,12 +3155,20 @@ TEST_CASE("gc.diskusagewindow")
 	SUBCASE("Find size window")
 	{
 		DiskUsageWindow Empty;
-		CHECK(Empty.FindTimepointThatRemoves(15u, 10000) == 10000);
 
-		CHECK(Stats.FindTimepointThatRemoves(15u, 40) == 21);
-		CHECK(Stats.FindTimepointThatRemoves(15u, 20) == 20);
-		CHECK(Stats.FindTimepointThatRemoves(100000u, 50) == 50);
-		CHECK(Stats.FindTimepointThatRemoves(100000u, 1000));
+		uint64_t FoundAmount = 0;
+
+		CHECK(Empty.FindTimepointThatRemoves(15u, 10000, FoundAmount) == 10000);
+		CHECK(FoundAmount == 0);
+
+		CHECK(Stats.FindTimepointThatRemoves(15u, 40, FoundAmount) == 21);
+		CHECK(FoundAmount == 20);
+		CHECK(Stats.FindTimepointThatRemoves(15u, 20, FoundAmount) == 11);
+		CHECK(FoundAmount == 10);
+		CHECK(Stats.FindTimepointThatRemoves(100000u, 50, FoundAmount) == 41);
+		CHECK(FoundAmount == 20);
+		CHECK(Stats.FindTimepointThatRemoves(100000u, 1000, FoundAmount) == 71);
+		CHECK(FoundAmount == 50);
 	}
 }