You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

PackIndexWriter.java 9.7KB

Support creating pack bitmap indexes in PackWriter. Update the PackWriter to support writing out pack bitmap indexes, a parallel ".bitmap" file to the ".pack" file. Bitmaps are selected at commits every 1 to 5,000 commits for each unique path from the start. The most recent 100 commits are all bitmapped. The next 19,000 commits have a bitmaps every 100 commits. The remaining commits have a bitmap every 5,000 commits. Commits with more than 1 parent are prefered over ones with 1 or less. Furthermore, previously computed bitmaps are reused, if the previous entry had the reuse flag set, which is set when the bitmap was placed at the max allowed distance. Bitmaps are used to speed up the counting phase when packing, for requests that are not shallow. The PackWriterBitmapWalker uses a RevFilter to proactively mark commits with RevFlag.SEEN, when they appear in a bitmap. The walker produces the full closure of reachable ObjectIds, given the collection of starting ObjectIds. For fetch request, two ObjectWalks are executed to compute the ObjectIds reachable from the haves and from the wants. The ObjectIds needed to be written are determined by taking all the resulting wants AND NOT the haves. For clone requests, we get cached pack support for "free" since it is possible to determine if all of the ObjectIds in a pack file are included in the resulting list of ObjectIds to write. On my machine, the best times for clones and fetches of the linux kernel repository (with about 2.6M objects and 300K commits) are tabulated below: Operation Index V2 Index VE003 Clone 37530ms (524.06 MiB) 82ms (524.06 MiB) Fetch (1 commit back) 75ms 107ms Fetch (10 commits back) 456ms (269.51 KiB) 341ms (265.19 KiB) Fetch (100 commits back) 449ms (269.91 KiB) 337ms (267.28 KiB) Fetch (1000 commits back) 2229ms ( 14.75 MiB) 189ms ( 14.42 MiB) Fetch (10000 commits back) 2177ms ( 16.30 MiB) 254ms ( 15.88 MiB) Fetch (100000 commits back) 14340ms (185.83 MiB) 1655ms (189.39 MiB) Change-Id: Icdb0cdd66ff168917fb9ef17b96093990cc6a98d
11 years ago
Support creating pack bitmap indexes in PackWriter. Update the PackWriter to support writing out pack bitmap indexes, a parallel ".bitmap" file to the ".pack" file. Bitmaps are selected at commits every 1 to 5,000 commits for each unique path from the start. The most recent 100 commits are all bitmapped. The next 19,000 commits have a bitmaps every 100 commits. The remaining commits have a bitmap every 5,000 commits. Commits with more than 1 parent are prefered over ones with 1 or less. Furthermore, previously computed bitmaps are reused, if the previous entry had the reuse flag set, which is set when the bitmap was placed at the max allowed distance. Bitmaps are used to speed up the counting phase when packing, for requests that are not shallow. The PackWriterBitmapWalker uses a RevFilter to proactively mark commits with RevFlag.SEEN, when they appear in a bitmap. The walker produces the full closure of reachable ObjectIds, given the collection of starting ObjectIds. For fetch request, two ObjectWalks are executed to compute the ObjectIds reachable from the haves and from the wants. The ObjectIds needed to be written are determined by taking all the resulting wants AND NOT the haves. For clone requests, we get cached pack support for "free" since it is possible to determine if all of the ObjectIds in a pack file are included in the resulting list of ObjectIds to write. On my machine, the best times for clones and fetches of the linux kernel repository (with about 2.6M objects and 300K commits) are tabulated below: Operation Index V2 Index VE003 Clone 37530ms (524.06 MiB) 82ms (524.06 MiB) Fetch (1 commit back) 75ms 107ms Fetch (10 commits back) 456ms (269.51 KiB) 341ms (265.19 KiB) Fetch (100 commits back) 449ms (269.91 KiB) 337ms (267.28 KiB) Fetch (1000 commits back) 2229ms ( 14.75 MiB) 189ms ( 14.42 MiB) Fetch (10000 commits back) 2177ms ( 16.30 MiB) 254ms ( 15.88 MiB) Fetch (100000 commits back) 14340ms (185.83 MiB) 1655ms (189.39 MiB) Change-Id: Icdb0cdd66ff168917fb9ef17b96093990cc6a98d
11 years ago
Support creating pack bitmap indexes in PackWriter. Update the PackWriter to support writing out pack bitmap indexes, a parallel ".bitmap" file to the ".pack" file. Bitmaps are selected at commits every 1 to 5,000 commits for each unique path from the start. The most recent 100 commits are all bitmapped. The next 19,000 commits have a bitmaps every 100 commits. The remaining commits have a bitmap every 5,000 commits. Commits with more than 1 parent are prefered over ones with 1 or less. Furthermore, previously computed bitmaps are reused, if the previous entry had the reuse flag set, which is set when the bitmap was placed at the max allowed distance. Bitmaps are used to speed up the counting phase when packing, for requests that are not shallow. The PackWriterBitmapWalker uses a RevFilter to proactively mark commits with RevFlag.SEEN, when they appear in a bitmap. The walker produces the full closure of reachable ObjectIds, given the collection of starting ObjectIds. For fetch request, two ObjectWalks are executed to compute the ObjectIds reachable from the haves and from the wants. The ObjectIds needed to be written are determined by taking all the resulting wants AND NOT the haves. For clone requests, we get cached pack support for "free" since it is possible to determine if all of the ObjectIds in a pack file are included in the resulting list of ObjectIds to write. On my machine, the best times for clones and fetches of the linux kernel repository (with about 2.6M objects and 300K commits) are tabulated below: Operation Index V2 Index VE003 Clone 37530ms (524.06 MiB) 82ms (524.06 MiB) Fetch (1 commit back) 75ms 107ms Fetch (10 commits back) 456ms (269.51 KiB) 341ms (265.19 KiB) Fetch (100 commits back) 449ms (269.91 KiB) 337ms (267.28 KiB) Fetch (1000 commits back) 2229ms ( 14.75 MiB) 189ms ( 14.42 MiB) Fetch (10000 commits back) 2177ms ( 16.30 MiB) 254ms ( 15.88 MiB) Fetch (100000 commits back) 14340ms (185.83 MiB) 1655ms (189.39 MiB) Change-Id: Icdb0cdd66ff168917fb9ef17b96093990cc6a98d
11 years ago
Support creating pack bitmap indexes in PackWriter. Update the PackWriter to support writing out pack bitmap indexes, a parallel ".bitmap" file to the ".pack" file. Bitmaps are selected at commits every 1 to 5,000 commits for each unique path from the start. The most recent 100 commits are all bitmapped. The next 19,000 commits have a bitmaps every 100 commits. The remaining commits have a bitmap every 5,000 commits. Commits with more than 1 parent are prefered over ones with 1 or less. Furthermore, previously computed bitmaps are reused, if the previous entry had the reuse flag set, which is set when the bitmap was placed at the max allowed distance. Bitmaps are used to speed up the counting phase when packing, for requests that are not shallow. The PackWriterBitmapWalker uses a RevFilter to proactively mark commits with RevFlag.SEEN, when they appear in a bitmap. The walker produces the full closure of reachable ObjectIds, given the collection of starting ObjectIds. For fetch request, two ObjectWalks are executed to compute the ObjectIds reachable from the haves and from the wants. The ObjectIds needed to be written are determined by taking all the resulting wants AND NOT the haves. For clone requests, we get cached pack support for "free" since it is possible to determine if all of the ObjectIds in a pack file are included in the resulting list of ObjectIds to write. On my machine, the best times for clones and fetches of the linux kernel repository (with about 2.6M objects and 300K commits) are tabulated below: Operation Index V2 Index VE003 Clone 37530ms (524.06 MiB) 82ms (524.06 MiB) Fetch (1 commit back) 75ms 107ms Fetch (10 commits back) 456ms (269.51 KiB) 341ms (265.19 KiB) Fetch (100 commits back) 449ms (269.91 KiB) 337ms (267.28 KiB) Fetch (1000 commits back) 2229ms ( 14.75 MiB) 189ms ( 14.42 MiB) Fetch (10000 commits back) 2177ms ( 16.30 MiB) 254ms ( 15.88 MiB) Fetch (100000 commits back) 14340ms (185.83 MiB) 1655ms (189.39 MiB) Change-Id: Icdb0cdd66ff168917fb9ef17b96093990cc6a98d
11 years ago
Support creating pack bitmap indexes in PackWriter. Update the PackWriter to support writing out pack bitmap indexes, a parallel ".bitmap" file to the ".pack" file. Bitmaps are selected at commits every 1 to 5,000 commits for each unique path from the start. The most recent 100 commits are all bitmapped. The next 19,000 commits have a bitmaps every 100 commits. The remaining commits have a bitmap every 5,000 commits. Commits with more than 1 parent are prefered over ones with 1 or less. Furthermore, previously computed bitmaps are reused, if the previous entry had the reuse flag set, which is set when the bitmap was placed at the max allowed distance. Bitmaps are used to speed up the counting phase when packing, for requests that are not shallow. The PackWriterBitmapWalker uses a RevFilter to proactively mark commits with RevFlag.SEEN, when they appear in a bitmap. The walker produces the full closure of reachable ObjectIds, given the collection of starting ObjectIds. For fetch request, two ObjectWalks are executed to compute the ObjectIds reachable from the haves and from the wants. The ObjectIds needed to be written are determined by taking all the resulting wants AND NOT the haves. For clone requests, we get cached pack support for "free" since it is possible to determine if all of the ObjectIds in a pack file are included in the resulting list of ObjectIds to write. On my machine, the best times for clones and fetches of the linux kernel repository (with about 2.6M objects and 300K commits) are tabulated below: Operation Index V2 Index VE003 Clone 37530ms (524.06 MiB) 82ms (524.06 MiB) Fetch (1 commit back) 75ms 107ms Fetch (10 commits back) 456ms (269.51 KiB) 341ms (265.19 KiB) Fetch (100 commits back) 449ms (269.91 KiB) 337ms (267.28 KiB) Fetch (1000 commits back) 2229ms ( 14.75 MiB) 189ms ( 14.42 MiB) Fetch (10000 commits back) 2177ms ( 16.30 MiB) 254ms ( 15.88 MiB) Fetch (100000 commits back) 14340ms (185.83 MiB) 1655ms (189.39 MiB) Change-Id: Icdb0cdd66ff168917fb9ef17b96093990cc6a98d
11 years ago
Support creating pack bitmap indexes in PackWriter. Update the PackWriter to support writing out pack bitmap indexes, a parallel ".bitmap" file to the ".pack" file. Bitmaps are selected at commits every 1 to 5,000 commits for each unique path from the start. The most recent 100 commits are all bitmapped. The next 19,000 commits have a bitmaps every 100 commits. The remaining commits have a bitmap every 5,000 commits. Commits with more than 1 parent are prefered over ones with 1 or less. Furthermore, previously computed bitmaps are reused, if the previous entry had the reuse flag set, which is set when the bitmap was placed at the max allowed distance. Bitmaps are used to speed up the counting phase when packing, for requests that are not shallow. The PackWriterBitmapWalker uses a RevFilter to proactively mark commits with RevFlag.SEEN, when they appear in a bitmap. The walker produces the full closure of reachable ObjectIds, given the collection of starting ObjectIds. For fetch request, two ObjectWalks are executed to compute the ObjectIds reachable from the haves and from the wants. The ObjectIds needed to be written are determined by taking all the resulting wants AND NOT the haves. For clone requests, we get cached pack support for "free" since it is possible to determine if all of the ObjectIds in a pack file are included in the resulting list of ObjectIds to write. On my machine, the best times for clones and fetches of the linux kernel repository (with about 2.6M objects and 300K commits) are tabulated below: Operation Index V2 Index VE003 Clone 37530ms (524.06 MiB) 82ms (524.06 MiB) Fetch (1 commit back) 75ms 107ms Fetch (10 commits back) 456ms (269.51 KiB) 341ms (265.19 KiB) Fetch (100 commits back) 449ms (269.91 KiB) 337ms (267.28 KiB) Fetch (1000 commits back) 2229ms ( 14.75 MiB) 189ms ( 14.42 MiB) Fetch (10000 commits back) 2177ms ( 16.30 MiB) 254ms ( 15.88 MiB) Fetch (100000 commits back) 14340ms (185.83 MiB) 1655ms (189.39 MiB) Change-Id: Icdb0cdd66ff168917fb9ef17b96093990cc6a98d
11 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267
  1. /*
  2. * Copyright (C) 2008, Robin Rosenberg <robin.rosenberg@dewire.com>
  3. * Copyright (C) 2008, Shawn O. Pearce <spearce@spearce.org> and others
  4. *
  5. * This program and the accompanying materials are made available under the
  6. * terms of the Eclipse Distribution License v. 1.0 which is available at
  7. * https://www.eclipse.org/org/documents/edl-v10.php.
  8. *
  9. * SPDX-License-Identifier: BSD-3-Clause
  10. */
  11. package org.eclipse.jgit.internal.storage.file;
  12. import java.io.BufferedOutputStream;
  13. import java.io.IOException;
  14. import java.io.OutputStream;
  15. import java.security.DigestOutputStream;
  16. import java.text.MessageFormat;
  17. import java.util.List;
  18. import org.eclipse.jgit.internal.JGitText;
  19. import org.eclipse.jgit.lib.Constants;
  20. import org.eclipse.jgit.transport.PackedObjectInfo;
  21. import org.eclipse.jgit.util.NB;
  22. /**
  23. * Creates a table of contents to support random access by
  24. * {@link org.eclipse.jgit.internal.storage.file.Pack}.
  25. * <p>
  26. * Pack index files (the <code>.idx</code> suffix in a pack file pair) provides
  27. * random access to any object in the pack by associating an ObjectId to the
  28. * byte offset within the pack where the object's data can be read.
  29. */
  30. public abstract class PackIndexWriter {
  31. /** Magic constant indicating post-version 1 format. */
  32. protected static final byte[] TOC = { -1, 't', 'O', 'c' };
  33. /**
  34. * Create a new writer for the oldest (most widely understood) format.
  35. * <p>
  36. * This method selects an index format that can accurate describe the
  37. * supplied objects and that will be the most compatible format with older
  38. * Git implementations.
  39. * <p>
  40. * Index version 1 is widely recognized by all Git implementations, but
  41. * index version 2 (and later) is not as well recognized as it was
  42. * introduced more than a year later. Index version 1 can only be used if
  43. * the resulting pack file is under 4 gigabytes in size; packs larger than
  44. * that limit must use index version 2.
  45. *
  46. * @param dst
  47. * the stream the index data will be written to. If not already
  48. * buffered it will be automatically wrapped in a buffered
  49. * stream. Callers are always responsible for closing the stream.
  50. * @param objs
  51. * the objects the caller needs to store in the index. Entries
  52. * will be examined until a format can be conclusively selected.
  53. * @return a new writer to output an index file of the requested format to
  54. * the supplied stream.
  55. * @throws java.lang.IllegalArgumentException
  56. * no recognized pack index version can support the supplied
  57. * objects. This is likely a bug in the implementation.
  58. * @see #oldestPossibleFormat(List)
  59. */
  60. public static PackIndexWriter createOldestPossible(final OutputStream dst,
  61. final List<? extends PackedObjectInfo> objs) {
  62. return createVersion(dst, oldestPossibleFormat(objs));
  63. }
  64. /**
  65. * Return the oldest (most widely understood) index format.
  66. * <p>
  67. * This method selects an index format that can accurate describe the
  68. * supplied objects and that will be the most compatible format with older
  69. * Git implementations.
  70. * <p>
  71. * Index version 1 is widely recognized by all Git implementations, but
  72. * index version 2 (and later) is not as well recognized as it was
  73. * introduced more than a year later. Index version 1 can only be used if
  74. * the resulting pack file is under 4 gigabytes in size; packs larger than
  75. * that limit must use index version 2.
  76. *
  77. * @param objs
  78. * the objects the caller needs to store in the index. Entries
  79. * will be examined until a format can be conclusively selected.
  80. * @return the index format.
  81. * @throws java.lang.IllegalArgumentException
  82. * no recognized pack index version can support the supplied
  83. * objects. This is likely a bug in the implementation.
  84. */
  85. public static int oldestPossibleFormat(
  86. final List<? extends PackedObjectInfo> objs) {
  87. for (PackedObjectInfo oe : objs) {
  88. if (!PackIndexWriterV1.canStore(oe))
  89. return 2;
  90. }
  91. return 1;
  92. }
  93. /**
  94. * Create a new writer instance for a specific index format version.
  95. *
  96. * @param dst
  97. * the stream the index data will be written to. If not already
  98. * buffered it will be automatically wrapped in a buffered
  99. * stream. Callers are always responsible for closing the stream.
  100. * @param version
  101. * index format version number required by the caller. Exactly
  102. * this formatted version will be written.
  103. * @return a new writer to output an index file of the requested format to
  104. * the supplied stream.
  105. * @throws java.lang.IllegalArgumentException
  106. * the version requested is not supported by this
  107. * implementation.
  108. */
  109. public static PackIndexWriter createVersion(final OutputStream dst,
  110. final int version) {
  111. switch (version) {
  112. case 1:
  113. return new PackIndexWriterV1(dst);
  114. case 2:
  115. return new PackIndexWriterV2(dst);
  116. default:
  117. throw new IllegalArgumentException(MessageFormat.format(
  118. JGitText.get().unsupportedPackIndexVersion,
  119. Integer.valueOf(version)));
  120. }
  121. }
  122. /** The index data stream we are responsible for creating. */
  123. protected final DigestOutputStream out;
  124. /** A temporary buffer for use during IO to {link #out}. */
  125. protected final byte[] tmp;
  126. /** The entries this writer must pack. */
  127. protected List<? extends PackedObjectInfo> entries;
  128. /** SHA-1 checksum for the entire pack data. */
  129. protected byte[] packChecksum;
  130. /**
  131. * Create a new writer instance.
  132. *
  133. * @param dst
  134. * the stream this instance outputs to. If not already buffered
  135. * it will be automatically wrapped in a buffered stream.
  136. */
  137. protected PackIndexWriter(OutputStream dst) {
  138. out = new DigestOutputStream(dst instanceof BufferedOutputStream ? dst
  139. : new BufferedOutputStream(dst),
  140. Constants.newMessageDigest());
  141. tmp = new byte[4 + Constants.OBJECT_ID_LENGTH];
  142. }
  143. /**
  144. * Write all object entries to the index stream.
  145. * <p>
  146. * After writing the stream passed to the factory is flushed but remains
  147. * open. Callers are always responsible for closing the output stream.
  148. *
  149. * @param toStore
  150. * sorted list of objects to store in the index. The caller must
  151. * have previously sorted the list using
  152. * {@link org.eclipse.jgit.transport.PackedObjectInfo}'s native
  153. * {@link java.lang.Comparable} implementation.
  154. * @param packDataChecksum
  155. * checksum signature of the entire pack data content. This is
  156. * traditionally the last 20 bytes of the pack file's own stream.
  157. * @throws java.io.IOException
  158. * an error occurred while writing to the output stream, or this
  159. * index format cannot store the object data supplied.
  160. */
  161. public void write(final List<? extends PackedObjectInfo> toStore,
  162. final byte[] packDataChecksum) throws IOException {
  163. entries = toStore;
  164. packChecksum = packDataChecksum;
  165. writeImpl();
  166. out.flush();
  167. }
  168. /**
  169. * Writes the index file to {@link #out}.
  170. * <p>
  171. * Implementations should go something like:
  172. *
  173. * <pre>
  174. * writeFanOutTable();
  175. * for (final PackedObjectInfo po : entries)
  176. * writeOneEntry(po);
  177. * writeChecksumFooter();
  178. * </pre>
  179. *
  180. * <p>
  181. * Where the logic for <code>writeOneEntry</code> is specific to the index
  182. * format in use. Additional headers/footers may be used if necessary and
  183. * the {@link #entries} collection may be iterated over more than once if
  184. * necessary. Implementors therefore have complete control over the data.
  185. *
  186. * @throws java.io.IOException
  187. * an error occurred while writing to the output stream, or this
  188. * index format cannot store the object data supplied.
  189. */
  190. protected abstract void writeImpl() throws IOException;
  191. /**
  192. * Output the version 2 (and later) TOC header, with version number.
  193. * <p>
  194. * Post version 1 all index files start with a TOC header that makes the
  195. * file an invalid version 1 file, and then includes the version number.
  196. * This header is necessary to recognize a version 1 from a version 2
  197. * formatted index.
  198. *
  199. * @param version
  200. * version number of this index format being written.
  201. * @throws java.io.IOException
  202. * an error occurred while writing to the output stream.
  203. */
  204. protected void writeTOC(int version) throws IOException {
  205. out.write(TOC);
  206. NB.encodeInt32(tmp, 0, version);
  207. out.write(tmp, 0, 4);
  208. }
  209. /**
  210. * Output the standard 256 entry first-level fan-out table.
  211. * <p>
  212. * The fan-out table is 4 KB in size, holding 256 32-bit unsigned integer
  213. * counts. Each count represents the number of objects within this index
  214. * whose {@link org.eclipse.jgit.lib.ObjectId#getFirstByte()} matches the
  215. * count's position in the fan-out table.
  216. *
  217. * @throws java.io.IOException
  218. * an error occurred while writing to the output stream.
  219. */
  220. protected void writeFanOutTable() throws IOException {
  221. final int[] fanout = new int[256];
  222. for (PackedObjectInfo po : entries)
  223. fanout[po.getFirstByte() & 0xff]++;
  224. for (int i = 1; i < 256; i++)
  225. fanout[i] += fanout[i - 1];
  226. for (int n : fanout) {
  227. NB.encodeInt32(tmp, 0, n);
  228. out.write(tmp, 0, 4);
  229. }
  230. }
  231. /**
  232. * Output the standard two-checksum index footer.
  233. * <p>
  234. * The standard footer contains two checksums (20 byte SHA-1 values):
  235. * <ol>
  236. * <li>Pack data checksum - taken from the last 20 bytes of the pack file.</li>
  237. * <li>Index data checksum - checksum of all index bytes written, including
  238. * the pack data checksum above.</li>
  239. * </ol>
  240. *
  241. * @throws java.io.IOException
  242. * an error occurred while writing to the output stream.
  243. */
  244. protected void writeChecksumFooter() throws IOException {
  245. out.write(packChecksum);
  246. out.on(false);
  247. out.write(out.getMessageDigest().digest());
  248. }
  249. }