X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=bluesky%2Flog.c;h=9ba75b471b1b4f4225c1c02fe3dc1d46c559acb2;hb=97517d1bbc647846003d1a49b6ae97c69d856b90;hp=55daf7a85939de2cf0502486980d9091d361c674;hpb=4fa8649924cb61b3d77ced461fe716a7ee18fc1d;p=bluesky.git diff --git a/bluesky/log.c b/bluesky/log.c index 55daf7a..9ba75b4 100644 --- a/bluesky/log.c +++ b/bluesky/log.c @@ -36,7 +36,7 @@ // Rough size limit for a log segment. This is not a firm limit and there are // no absolute guarantees on the size of a log segment. -#define LOG_SEGMENT_SIZE (1 << 23) +#define LOG_SEGMENT_SIZE (1 << 24) #define HEADER_MAGIC 0x676f4c0a #define FOOTER_MAGIC 0x2e435243 @@ -77,11 +77,14 @@ static void log_commit(BlueSkyLog *log) while (log->committed != NULL) { BlueSkyCloudLog *item = (BlueSkyCloudLog *)log->committed->data; g_mutex_lock(item->lock); + bluesky_cloudlog_stats_update(item, -1); item->pending_write &= ~CLOUDLOG_JOURNAL; item->location_flags |= CLOUDLOG_JOURNAL; + bluesky_cloudlog_stats_update(item, 1); g_cond_signal(item->cond); g_mutex_unlock(item->lock); log->committed = g_slist_delete_link(log->committed, log->committed); + bluesky_cloudlog_unref(item); batchsize++; } @@ -101,7 +104,7 @@ static gboolean log_open(BlueSkyLog *log) } while (log->fd < 0) { - g_snprintf(logname, sizeof(logname), "log-%08d", log->seq_num); + g_snprintf(logname, sizeof(logname), "journal-%08d", log->seq_num); log->fd = openat(log->dirfd, logname, O_CREAT|O_WRONLY|O_EXCL, 0600); if (log->fd < 0 && errno == EEXIST) { fprintf(stderr, "Log file %s already exists...\n", logname); @@ -153,10 +156,13 @@ static gpointer log_thread(gpointer d) if ((item->location_flags | item->pending_write) & CLOUDLOG_JOURNAL) { g_mutex_unlock(item->lock); bluesky_cloudlog_unref(item); + g_atomic_int_add(&item->data_lock_count, -1); continue; } + bluesky_cloudlog_stats_update(item, -1); item->pending_write |= CLOUDLOG_JOURNAL; + bluesky_cloudlog_stats_update(item, 1); struct log_header header; struct log_footer footer; @@ -198,7 +204,16 @@ static gpointer log_thread(gpointer d) offset += sizeof(header) + sizeof(footer) + item->data->len; + /* Replace the log item's string data with a memory-mapped copy of the + * data, now that it has been written to the log file. (Even if it + * isn't yet on disk, it should at least be in the page cache and so + * available to memory map.) */ + bluesky_string_unref(item->data); + item->data = NULL; + bluesky_cloudlog_fetch(item); + log->committed = g_slist_prepend(log->committed, item); + g_atomic_int_add(&item->data_lock_count, -1); g_mutex_unlock(item->lock); /* Force an if there are no other log items currently waiting to be @@ -235,6 +250,7 @@ BlueSkyLog *bluesky_log_new(const char *log_directory) void bluesky_log_item_submit(BlueSkyCloudLog *item, BlueSkyLog *log) { bluesky_cloudlog_ref(item); + g_atomic_int_add(&item->data_lock_count, 1); g_async_queue_push(log->queue, item); } @@ -257,72 +273,161 @@ void bluesky_log_finish_all(GList *log_items) * to it. */ static int page_size = 0; -BlueSkyRCStr *bluesky_log_map_object(BlueSkyLog *log, - int log_seq, int log_offset, int log_size) +void bluesky_cachefile_unref(BlueSkyCacheFile *cachefile) +{ + g_atomic_int_add(&cachefile->refcount, -1); +} + +static void cloudlog_fetch_complete(BlueSkyStoreAsync *async, + BlueSkyCacheFile *cachefile) +{ + g_print("Fetch of %s from cloud complete, status = %d\n", + async->key, async->result); + + if (async->result < 0) + return; + + g_mutex_lock(cachefile->lock); + char *pathname = g_strdup_printf("%s/%s", cachefile->log->log_directory, + cachefile->filename); + if (!g_file_set_contents(pathname, async->data->data, async->data->len, + NULL)) + { + g_print("Error writing out fetched file to cache!\n"); + } + g_free(pathname); + cachefile->fetching = FALSE; + cachefile->ready = TRUE; + bluesky_cachefile_unref(cachefile); + g_cond_broadcast(cachefile->cond); + g_mutex_unlock(cachefile->lock); +} + +/* Find the BlueSkyCacheFile object for the given journal or cloud log segment. + * Returns the object in the locked state and with a reference taken. */ +BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs, + int clouddir, int log_seq) { if (page_size == 0) { page_size = getpagesize(); } - BlueSkyMmap *map; + BlueSkyLog *log = fs->log; + + BlueSkyCacheFile *map; g_mutex_lock(log->mmap_lock); map = g_hash_table_lookup(log->mmap_cache, GINT_TO_POINTER(log_seq)); if (map == NULL) { - char logname[64]; - g_snprintf(logname, sizeof(logname), "log-%08d", log_seq); - int fd = openat(log->dirfd, logname, O_RDONLY); + char *logname; + + // A request for a local log file + if (clouddir < 0) { + logname = g_strdup_printf("journal-%08d", log_seq); + } else { + logname = g_strdup_printf("log-%08d-%08d", clouddir, log_seq); + } + + /* TODO: stat() call */ + map = g_new0(BlueSkyCacheFile, 1); + map->type = CLOUDLOG_JOURNAL; + map->lock = g_mutex_new(); + g_mutex_lock(map->lock); + map->cond = g_cond_new(); + map->filename = logname; + map->log_seq = log_seq; + map->log = log; + g_atomic_int_set(&map->mapcount, 0); + g_atomic_int_set(&map->refcount, 0); + + g_hash_table_insert(log->mmap_cache, GINT_TO_POINTER(log_seq), map); + + // If the log file is stored in the cloud, we may need to fetch it + if (clouddir >= 0) { + g_atomic_int_inc(&map->refcount); + map->fetching = TRUE; + g_print("Starting fetch of %s from cloud\n", logname); + BlueSkyStoreAsync *async = bluesky_store_async_new(fs->store); + async->op = STORE_OP_GET; + async->key = g_strdup(logname); + bluesky_store_async_add_notifier(async, + (GFunc)cloudlog_fetch_complete, + map); + bluesky_store_async_submit(async); + bluesky_store_async_unref(async); + } + } else { + g_mutex_lock(map->lock); + } + + g_mutex_unlock(log->mmap_lock); + g_atomic_int_inc(&map->refcount); + return map; +} + +BlueSkyRCStr *bluesky_log_map_object(BlueSkyFS *fs, int log_dir, + int log_seq, int log_offset, int log_size) +{ + if (page_size == 0) { + page_size = getpagesize(); + } + + BlueSkyLog *log = fs->log; + BlueSkyCacheFile *map = bluesky_cachefile_lookup(fs, log_dir, log_seq); + + if (map == NULL) { + return NULL; + } + + if (map->addr == NULL) { + while (!map->ready && map->fetching) { + g_print("Waiting for log segment to be fetched from cloud...\n"); + g_cond_wait(map->cond, map->lock); + } + + int fd = openat(log->dirfd, map->filename, O_RDONLY); if (fd < 0) { - fprintf(stderr, "Error opening logfile %s: %m\n", logname); - g_mutex_unlock(log->mmap_lock); + fprintf(stderr, "Error opening logfile %s: %m\n", map->filename); + bluesky_cachefile_unref(map); + g_mutex_unlock(map->lock); return NULL; } - map = g_new0(BlueSkyMmap, 1); - off_t length = lseek(fd, 0, SEEK_END); - map->log_seq = log_seq; map->addr = (const char *)mmap(NULL, length, PROT_READ, MAP_SHARED, fd, 0); map->len = length; - map->log = log; - g_atomic_int_set(&map->refcount, 0); - g_hash_table_insert(log->mmap_cache, GINT_TO_POINTER(log_seq), map); + g_print("Re-mapped log segment %d...\n", log_seq); + g_atomic_int_inc(&map->refcount); close(fd); } g_mutex_unlock(log->mmap_lock); - return bluesky_string_new_from_mmap(map, log_offset, log_size); + BlueSkyRCStr *str; + map->atime = bluesky_get_current_time(); + str = bluesky_string_new_from_mmap(map, log_offset, log_size); + bluesky_cachefile_unref(map); + g_mutex_unlock(map->lock); + return str; } -void bluesky_mmap_unref(BlueSkyMmap *mmap) +void bluesky_mmap_unref(BlueSkyCacheFile *mmap) { if (mmap == NULL) return; - if (g_atomic_int_dec_and_test(&mmap->refcount)) { - /* There is a potential race condition here: the BlueSkyLog contains a - * hash table of currently-existing BlueSkyMmap objects, which does not - * hold a reference. Some other thread might grab a new reference to - * this object after reading it from the hash table. So, before - * destruction we need to grab the lock for the hash table, then check - * the reference count again. If it is still zero, we can proceed with - * object destruction. */ - BlueSkyLog *log = mmap->log; - g_mutex_lock(log->mmap_lock); - if (g_atomic_int_get(&mmap->refcount) > 0) { - g_mutex_unlock(log->mmap_lock); - return; + if (g_atomic_int_dec_and_test(&mmap->mapcount)) { + g_mutex_lock(mmap->lock); + if (g_atomic_int_get(&mmap->mapcount) == 0) { + g_print("Unmapped log segment %d...\n", mmap->log_seq); + munmap((void *)mmap->addr, mmap->len); + mmap->addr = NULL; + g_atomic_int_add(&mmap->refcount, -1); } - - g_hash_table_remove(log->mmap_cache, GINT_TO_POINTER(mmap->log_seq)); - munmap((void *)mmap->addr, mmap->len); - g_free(mmap); - g_mutex_unlock(log->mmap_lock); + g_mutex_unlock(mmap->lock); } } -