Backward incompatible file format changes to enforce headers aligne t…

…o system page size

src/main/java/com/upserve/uppend/AppendOnlyStoreBuilder.java

@@ -1,5 +1,6 @@
 package com.upserve.uppend;
 
+import com.upserve.uppend.blobs.NativeIO;
 import com.upserve.uppend.metrics.*;
 
 public class AppendOnlyStoreBuilder extends FileStoreBuilder<AppendOnlyStoreBuilder> {
@@ -8,7 +9,7 @@ public class AppendOnlyStoreBuilder extends FileStoreBuilder<AppendOnlyStoreBuil
     private int blobsPerBlock = DEFAULT_BLOBS_PER_BLOCK;
 
     // Blob Cache Options
-    public static final int DEFAULT_BLOB_PAGE_SIZE = 4 * 1024 * 1024;
+    public static final int DEFAULT_BLOB_PAGE_SIZE =  NativeIO.pageSize * 1024;
     private int blobPageSize = DEFAULT_BLOB_PAGE_SIZE;
 
 
@@ -26,6 +27,14 @@ public AppendOnlyStoreBuilder withBlobsPerBlock(int blobsPerBlock) {
 
     // Blob Options
     public AppendOnlyStoreBuilder withBlobPageSize(int blobPageSize) {
+        if (blobPageSize % NativeIO.pageSize != 0) {
+            throw new IllegalArgumentException(
+                    String.format(
+                            "Illegal blobPageSize %d; Must be a multiple of the host system page size: %d",
+                            blobPageSize, NativeIO.pageSize
+                    )
+            );
+        }
         this.blobPageSize = blobPageSize;
         return this;
     }

src/main/java/com/upserve/uppend/BlockedLongs.java

@@ -25,7 +25,7 @@ public class BlockedLongs implements AutoCloseable, Flushable {
     private static final int PAGE_SIZE = 128 * 1024 * 1024; // allocate 128 MB chunks
     private static final int MAX_PAGES = 32 * 1024; // max 4 TB
 
-    static final int HEADER_BYTES = 128; // Currently 16 used for file size and append count
+    static final int HEADER_BYTES = NativeIO.pageSize; // Currently 16 used for file size and append count
     private static final int posBufPosition = 0;
     private static final int appendBufPosition = 8;
 
@@ -58,6 +58,15 @@ public class BlockedLongs implements AutoCloseable, Flushable {
             throw new IllegalArgumentException("null file");
         }
 
+        if (PAGE_SIZE % NativeIO.pageSize != 0) {
+            throw new IllegalArgumentException(
+                    String.format(
+                            "The BlockedLong PAGE SIZE %d is not a multiple of the system page size %d on this OS",
+                            PAGE_SIZE, NativeIO.pageSize
+                    )
+            );
+        }
+
         this.file = file;
         this.readOnly = readOnly;
         this.blockedLongMetricsAdders = blockedLongMetricsAdders;
@@ -131,13 +140,12 @@ else if (pos < HEADER_BYTES) {
 
         try {
             appendCountBuf = blocks.map(readOnly ? FileChannel.MapMode.READ_ONLY : FileChannel.MapMode.READ_WRITE, appendBufPosition, 8);
+            NativeIO.madvise(appendCountBuf, NativeIO.Advice.WillNeed);
         } catch (IOException e) {
             throw new UncheckedIOException("Unable to map pos buffer at in " + file, e);
         }
         initialAppendCount = appendCountBuf.getLong(0);
 
-
-
         posMem = new AtomicLong(pos);
     }
 

src/main/java/com/upserve/uppend/FileStoreBuilder.java

@@ -1,6 +1,7 @@
 package com.upserve.uppend;
 
 import com.codahale.metrics.MetricRegistry;
+import com.upserve.uppend.blobs.NativeIO;
 import com.upserve.uppend.metrics.*;
 
 import java.nio.file.Path;
@@ -10,11 +11,11 @@ public class FileStoreBuilder<T extends FileStoreBuilder<T>> {
     // Long lookup Cache Options
     public static final int DEFAULT_PARTITION_COUNT = 0;
     public static final int DEFAULT_LOOKUP_HASH_COUNT = 256;
-    public static final int DEFAULT_LOOKUP_PAGE_SIZE = 256 * 1024;
+    public static final int DEFAULT_LOOKUP_PAGE_SIZE =  NativeIO.pageSize * 64;
 
     public static final int TARGET_PRODUCTION_BUFFER_SIZE = Integer.MAX_VALUE;
 
-    public static final int DEFAULT_METADATA_PAGE_SIZE = 4096;
+    public static final int DEFAULT_METADATA_PAGE_SIZE = NativeIO.pageSize;
     public static final int DEFAULT_METADATA_TTL = 0; // Off by default!
 
     private String storeName = "";
@@ -56,12 +57,28 @@ public T withLongLookupHashCount(int longLookupHashCount) {
 
     @SuppressWarnings("unchecked")
     public T withLookupPageSize(int lookupPageSize) {
+        if (lookupPageSize % NativeIO.pageSize != 0) {
+            throw new IllegalArgumentException(
+                    String.format(
+                            "Illegal lookupPageSize %d; Must be a multiple of the host system page size: %d",
+                            lookupPageSize, NativeIO.pageSize
+                    )
+            );
+        }
         this.lookupPageSize = lookupPageSize;
         return (T) this;
     }
 
     @SuppressWarnings("unchecked")
     public T withMetadataPageSize(int metadataPageSize) {
+        if (metadataPageSize % NativeIO.pageSize != 0){
+            throw new IllegalArgumentException(
+                    String.format(
+                            "Illegal metadataPageSize %d; Must be a multiple of the host system page size: %d",
+                            metadataPageSize, NativeIO.pageSize
+                    )
+            );
+        }
         this.metadataPageSize = metadataPageSize;
         return (T) this;
     }

src/main/java/com/upserve/uppend/blobs/NativeIO.java

@@ -13,7 +13,7 @@ public class NativeIO {
     private static final Logger log = org.slf4j.LoggerFactory.getLogger(MethodHandles.lookup().lookupClass());
 
     private static final NativeC nativeC = LibraryLoader.create(NativeC.class).load("c");
-    static final int pageSize = nativeC.getpagesize(); // 4096 on most Linux
+    public static final int pageSize = nativeC.getpagesize(); // 4096 on most Linux
 
     public enum Advice {
         // These seem to be fairly stable https://github.com/torvalds/linux

src/main/java/com/upserve/uppend/blobs/VirtualPageFile.java

@@ -174,6 +174,10 @@ long nextAlignedPosition(long position, int lowBound, int highBound) {
         }
     }
 
+    int roundUpto(int size, int incriments) {
+        return (size + incriments - 1) & (-incriments);
+    }
+
     long getPosition(int virtualFileNumber) {
         if (readOnly) {
             return getHeaderVirtualFilePosition(virtualFileNumber);
@@ -332,7 +336,10 @@ public VirtualPageFile(Path filePath, int virtualFiles, int pageSize, int target
             throw new UncheckedIOException("Unable to read, write, map or get the size of " + getFilePath(), e);
         }
 
-        totalHeaderSize = headerSize + tableSize + SELF_DESCRIBING_HEADER_SIZE + PAGE_TABLE_BLOCK_LOCATION_HEADER_SIZE;
+        totalHeaderSize = roundUpto(
+                headerSize + tableSize + SELF_DESCRIBING_HEADER_SIZE + PAGE_TABLE_BLOCK_LOCATION_HEADER_SIZE,
+                NativeIO.pageSize
+        );
 
         try {
             headerBuffer = channel.map(mapMode, SELF_DESCRIBING_HEADER_SIZE + PAGE_TABLE_BLOCK_LOCATION_HEADER_SIZE, headerSize);

src/test/java/com/upserve/uppend/BlockedLongsTest.java

@@ -85,7 +85,7 @@ public void testAppendAtNonStartingBlock() throws Exception {
             v.append(pos1, i);
         }
         int blockSize = 16 + 10 * 8; // mirrors BlockedLongs.blockSize
-        v.append(blockSize * 2, 21);
+        v.append(v.HEADER_BYTES + blockSize * 2, 21);
     }
 
     @Test(expected = IllegalStateException.class)

src/test/java/com/upserve/uppend/TestHelper.java

@@ -1,5 +1,6 @@
 package com.upserve.uppend;
 
+import com.upserve.uppend.blobs.NativeIO;
 import org.junit.*;
 import org.slf4j.LoggerFactory;
 import java.io.*;
@@ -88,19 +89,19 @@ public static AppendOnlyStoreBuilder getDefaultAppendStoreTestBuilder() {
     public static AppendOnlyStoreBuilder getDefaultAppendStoreTestBuilder(ExecutorService testService) {
         return new AppendOnlyStoreBuilder()
                 .withStoreName("test")
-                .withBlobPageSize(64 * 1024)
+                .withBlobPageSize(8 * NativeIO.pageSize)
                 .withBlobsPerBlock(30)
                 .withTargetBufferSize(16*1024*1024)
                 .withLongLookupHashCount(16)
-                .withLookupPageSize(16 * 1024)
+                .withLookupPageSize(4 * NativeIO.pageSize)
                 .withMetadataTTL(0);
     }
 
     public static CounterStoreBuilder getDefaultCounterStoreTestBuilder() {
         return new CounterStoreBuilder()
                 .withStoreName("test")
                 .withTargetBufferSize(16*1024*1024)
-                .withMetadataPageSize(1024)
+                .withMetadataPageSize(NativeIO.pageSize)
                 .withLongLookupHashCount(16)
                 .withLookupPageSize(16 * 1024);
     }

src/test/java/com/upserve/uppend/blobs/VirtualPageFileTest.java

@@ -81,13 +81,13 @@ public void testReadOnlyTruncation() throws IOException {
 
         page = instance.getExistingPage(5,0);
 
-        assertEquals(313016, instance.getFileSize());
+        assertEquals(315392, instance.getFileSize());
 
         instance.close();
 
         // We can open the file in read only after truncation
         VirtualPageFile roInstance = new VirtualPageFile(path, 36, 1024, 16384, true);
-        assertEquals(313016, roInstance.getFileSize());
+        assertEquals(315392, roInstance.getFileSize());
 
         page = roInstance.getExistingPage(5,0);
 
@@ -97,20 +97,20 @@ public void testReadOnlyTruncation() throws IOException {
 
         // Make a new page - check that file is extended again.
         instance = new VirtualPageFile(path, 36, 1024, 16384, false);
-        assertEquals(313016L, instance.getFileSize());
+        assertEquals(315392L, instance.getFileSize());
 
         page = instance.getOrCreatePage(6,0);
         page.put(6, "def".getBytes(), 0);
         page.put(900, "ghi".getBytes(), 0);
 
         page = roInstance.getExistingPage(6,0);
-        assertEquals(313016L, roInstance.getFileSize());
+        assertEquals(315392L, roInstance.getFileSize());
         page.get(6, bytes, 0);
         assertArrayEquals("def".getBytes(), bytes);
 
         instance.close();
 
-        assertEquals(298680L, roInstance.getFileSize());
+        assertEquals(301056L, roInstance.getFileSize());
 
         page.get(900, bytes, 0);
         assertArrayEquals("ghi".getBytes(), bytes);