From bc2dbc3601b8281777a4ff9002645aaf9cbd6b62 Mon Sep 17 00:00:00 2001
From: Gavin Halliday <gavin.halliday@lexisnexis.com>
Date: Thu, 28 Nov 2024 11:33:05 +0000
Subject: [PATCH] HPCC-33045 Add an option to check size consistency within the
 write sync margin

Signed-off-by: Gavin Halliday <gavin.halliday@lexisnexis.com>
---
 dali/base/dadfs.cpp | 46 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 46 insertions(+)

diff --git a/dali/base/dadfs.cpp b/dali/base/dadfs.cpp
index 005767ac263..25c85376af5 100644
--- a/dali/base/dadfs.cpp
+++ b/dali/base/dadfs.cpp
@@ -90,6 +90,32 @@ enum MDFSRequestKind
     MDFS_MAX
 };
 
+
+//--------------------------------------------------------------------------------------------------------------------
+
+static unsigned dadfsHookId = 0;
+static bool expertVerifyWriteSyncSizes = false;
+
+//Add any code for caching global configuration here...
+MODULE_INIT(INIT_PRIORITY_STANDARD)
+{
+    auto updateFunc = [&](const IPropertyTree *oldComponentConfiguration, const IPropertyTree *oldGlobalConfiguration)
+    {
+        Owned<IPropertyTree> componentConfig = getComponentConfig();
+        Owned<IPropertyTree> globalConfig = getGlobalConfig();
+        expertVerifyWriteSyncSizes = componentConfig->getPropBool("expert/@verifyWriteSyncSizes", globalConfig->getPropBool("expert/@verifyWriteSyncSizes", false));
+    };
+    dadfsHookId = installConfigUpdateHook(updateFunc, true);
+    return true;
+}
+
+MODULE_EXIT()
+{
+    removeConfigUpdateHook(dadfsHookId);
+}
+
+//--------------------------------------------------------------------------------------------------------------------
+
 // Mutex for physical operations (remove/rename)
 static CriticalSection physicalChange;
 
@@ -3813,13 +3839,33 @@ protected: friend class CDistributedFilePart;
                 if (unlikely(elapsedMs < marginMs))
                 {
                     LOG(MCuserProgress, "Delaying access to %s on %s for %ums to ensure write sync", queryLogicalName(), name, (unsigned)(marginMs - elapsedMs));
+                    if (expertVerifyWriteSyncSizes)
+                        checkSizeConsistency("Before write sync delay");
                     MilliSleep(marginMs - elapsedMs);
+                    if (expertVerifyWriteSyncSizes)
+                        checkSizeConsistency("After write sync delay");
                     now = 0; // re-evaluate now - unlikely to actually happen
                 }
             }
         }
     }
 
+    void checkSizeConsistency(const char * when)
+    {
+        auto checkPartSize = [this,when](unsigned idx) -> void
+        {
+            IDistributedFilePart &part = queryPart(idx);
+            offset_t size = part.getDiskSize(false, false);
+            if (size != (offset_t)-1)
+            {
+                offset_t physicalSize = part.getDiskSize(true, true);
+                if (size != physicalSize)
+                    OWARNLOG("%s: Part %d of file %s is inconsistent: logical size = %" I64F "d, physical size = %" I64F "d", when, idx, queryLogicalName(), size, physicalSize);
+            }
+        };
+        asyncFor(numParts(), checkPartSize);
+    }
+
     bool hasDirPerPart() const
     {
         return FileDescriptorFlags::none != (fileFlags & FileDescriptorFlags::dirperpart);