X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=49d8c329351ee92d55ddcbf5e006a210850c96e7;hb=6bb973b965a30832c3e2d9f6a24e80d3309ef89e;hp=588a907e0337a774f58713a197bb43601e733507;hpb=edcae8181571b57b058cb2249cb84b01df73d52a;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index 588a907..49d8c32 100644 --- a/scandir.cc +++ b/scandir.cc @@ -25,6 +25,7 @@ #include #include "localdb.h" +#include "metadata.h" #include "store.h" #include "sha1.h" #include "statcache.h" @@ -44,13 +45,12 @@ using std::ostream; static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION); static TarSegmentStore *tss = NULL; +static MetadataWriter *metawriter = NULL; /* Buffer for holding a single block of data read from a file. */ static const size_t LBS_BLOCK_SIZE = 1024 * 1024; static char *block_buf; -static const size_t LBS_METADATA_BLOCK_SIZE = 65536; - /* Local database, which tracks objects written in this and previous * invocations to help in creating incremental snapshots. */ LocalDb *db; @@ -59,13 +59,6 @@ LocalDb *db; * skipping files which have not changed. */ StatCache *statcache; -/* Contents of the root object. This will contain a set of indirect links to - * the metadata objects. */ -std::ostringstream metadata_root; - -/* Buffer for building up metadata. */ -std::ostringstream metadata; - /* Keep track of all segments which are needed to reconstruct the snapshot. */ std::set segment_list; @@ -78,28 +71,11 @@ std::list searches; // Directories we don't want to save, but bool relative_paths = true; -/* Ensure contents of metadata are flushed to an object. */ -void metadata_flush() +/* Ensure that the given segment is listed as a dependency of the current + * snapshot. */ +void add_segment(const string& segment) { - string m = metadata.str(); - if (m.size() == 0) - return; - - /* Write current metadata information to a new object. */ - LbsObject *meta = new LbsObject; - meta->set_group("metadata"); - meta->set_data(m.data(), m.size()); - meta->write(tss); - meta->checksum(); - - /* Write a reference to this block in the root. */ - ObjectReference ref = meta->get_ref(); - metadata_root << "@" << ref.to_string() << "\n"; - segment_list.insert(ref.get_segment()); - - delete meta; - - metadata.str(""); + segment_list.insert(segment); } /* Read data from a file descriptor and return the amount of data read. A @@ -200,6 +176,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, // Store a copy of the object if one does not yet exist if (ref.get_segment().size() == 0) { LbsObject *o = new LbsObject; + int object_group; /* We might still have seen this checksum before, if the object * was stored at some time in the past, but we have decided to @@ -212,8 +189,15 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, * Additionally, keep track of the age of the data by looking * up the age of the block which was expired and using that * instead of the current time. */ - if (db->IsOldObject(block_csum, bytes, &block_age)) { - o->set_group("compacted"); + if (db->IsOldObject(block_csum, bytes, + &block_age, &object_group)) { + if (object_group == 0) { + o->set_group("data"); + } else { + char group[32]; + sprintf(group, "compacted-%d", object_group); + o->set_group(group); + } if (status == NULL) status = "partial"; } else { @@ -340,7 +324,7 @@ void dump_inode(const string& path, // Path within snapshot fprintf(stderr, "error reading symlink: %m\n"); } else if (len <= stat_buf.st_size) { buf[len] = '\0'; - file_info["contents"] = uri_encode(buf); + file_info["target"] = uri_encode(buf); } else if (len > stat_buf.st_size) { fprintf(stderr, "error reading symlink: name truncated\n"); } @@ -348,7 +332,7 @@ void dump_inode(const string& path, // Path within snapshot delete[] buf; break; case S_IFREG: - inode_type = '-'; + inode_type = 'f'; file_size = dumpfile(fd, file_info, path, stat_buf); file_info["size"] = encode_int(file_size); @@ -373,13 +357,7 @@ void dump_inode(const string& path, // Path within snapshot file_info["type"] = string(1, inode_type); - metadata << "name: " << uri_encode(path) << "\n"; - dict_output(metadata, file_info); - metadata << "\n"; - - // Break apart metadata listing if it becomes too large. - if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE) - metadata_flush(); + metawriter->add(path, file_info); } void scanfile(const string& path, bool include) @@ -705,6 +683,8 @@ int main(int argc, char *argv[]) tss = new TarSegmentStore(backup_dest, db); + metawriter = new MetadataWriter(tss); + /* Initialize the stat cache, for skipping over unchanged files. */ statcache = new StatCache; statcache->Open(localdb_dir.c_str(), desc_buf, @@ -712,22 +692,15 @@ int main(int argc, char *argv[]) scanfile(".", false); - metadata_flush(); - const string md = metadata_root.str(); - - LbsObject *root = new LbsObject; - root->set_group("metadata"); - root->set_data(md.data(), md.size()); - root->write(tss); - root->checksum(); - segment_list.insert(root->get_ref().get_segment()); - - string backup_root = root->get_ref().to_string(); - delete root; + ObjectReference root_ref = metawriter->close(); + add_segment(root_ref.get_segment()); + string backup_root = root_ref.to_string(); statcache->Close(); delete statcache; + delete metawriter; + tss->sync(); tss->dump_stats(); delete tss; @@ -791,7 +764,7 @@ int main(int argc, char *argv[]) } FILE *descriptor = fdopen(descriptor_fd, "w"); - fprintf(descriptor, "Format: LBS Snapshot v0.2\n"); + fprintf(descriptor, "Format: LBS Snapshot v0.6\n"); fprintf(descriptor, "Producer: LBS %s\n", lbs_version); strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf); fprintf(descriptor, "Date: %s\n", desc_buf);