X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=bluesky%2Fcache.c;h=cdd5c65e30f65975b6bfe3c1991b19ae1952b98e;hb=7f01069131555a0e9ab332e578a9eb2815e1e12d;hp=58b7cd6f1ac79c6b67af07fc4ece5979f872312d;hpb=e49fed25d000a51b316b480c1bfb1b00c1f5d345;p=bluesky.git diff --git a/bluesky/cache.c b/bluesky/cache.c index 58b7cd6..cdd5c65 100644 --- a/bluesky/cache.c +++ b/bluesky/cache.c @@ -6,14 +6,25 @@ * TODO: Licensing */ +#define _GNU_SOURCE +#define _ATFILE_SOURCE + +#include #include +#include #include #include +#include #include +#include +#include +#include +#include #include "bluesky-private.h" #define WRITEBACK_DELAY (20 * 1000000) +#define CACHE_DROP_DELAY (20 * 1000000) /* Filesystem caching and cache coherency. There are actually a couple of * different tasks that are performed here: @@ -22,40 +33,13 @@ * - Writing batches of data to the cloud. */ -#if 0 -static void writeback_complete(gpointer a, gpointer i) -{ - BlueSkyInode *inode = (BlueSkyInode *)i; - - if (bluesky_verbose) { - g_log("bluesky/flushd", G_LOG_LEVEL_DEBUG, - "Writeback for inode %"PRIu64" complete", inode->inum); - } - - g_mutex_lock(inode->lock); - - inode->change_commit = inode->change_pending; - inode->change_pending = 0; - if (inode->change_count == inode->change_commit) { - /* If inode is no longer dirty... */ - inode->change_time = 0; - g_mutex_lock(inode->fs->lock); - bluesky_list_unlink(&inode->fs->dirty_list, inode->dirty_list); - inode->dirty_list = NULL; - g_mutex_unlock(inode->fs->lock); - } - - g_mutex_unlock(inode->lock); -} -#endif - static void flushd_dirty_inode(BlueSkyInode *inode) { BlueSkyFS *fs = inode->fs; g_mutex_lock(fs->lock); - bluesky_list_unlink(&fs->dirty_list, inode->dirty_list); - inode->dirty_list = NULL; + bluesky_list_unlink(&fs->unlogged_list, inode->unlogged_list); + inode->unlogged_list = NULL; g_mutex_unlock(fs->lock); /* Inode is clean; nothing to do. */ @@ -70,6 +54,17 @@ static void flushd_dirty_inode(BlueSkyInode *inode) bluesky_inode_start_sync(inode); } +/* Check whether memory usage may have dropped below critical thresholds for + * waking up waiting threads. */ +void flushd_check_wakeup(BlueSkyFS *fs) +{ + int dirty = g_atomic_int_get(&fs->cache_dirty); + dirty += g_atomic_int_get(&fs->cache_log_dirty); + + if (dirty <= bluesky_watermark_high_dirty) + g_cond_broadcast(fs->flushd_cond); +} + /* Try to flush dirty data to disk, either due to memory pressure or due to * timeouts. */ static void flushd_dirty(BlueSkyFS *fs) @@ -79,9 +74,9 @@ static void flushd_dirty(BlueSkyFS *fs) while (1) { BlueSkyInode *inode; - if (fs->dirty_list.prev == NULL) + if (fs->unlogged_list.prev == NULL) break; - inode = fs->dirty_list.prev->data; + inode = fs->unlogged_list.prev->data; if (bluesky_verbose) { g_log("bluesky/flushd", G_LOG_LEVEL_DEBUG, @@ -108,20 +103,143 @@ static void flushd_dirty(BlueSkyFS *fs) bluesky_inode_unref(inode); g_mutex_lock(fs->lock); + flushd_check_wakeup(fs); } + g_cond_broadcast(fs->flushd_cond); + g_mutex_unlock(fs->lock); } +/* Try to flush dirty data to the cloud. This will take a snapshot of the + * entire filesystem (though only point-in-time consistent for isolated inodes + * and not the filesystem as a whole) and ensure all data is written to the + * cloud. When the write completes, we will allow old journal segments (those + * that were fully written _before_ the snapshot process started) to be garbage + * collected. Newer journal segments can't be collected yet since they may + * still contain data which has not been written persistently to the cloud. + * + * Note that some of this code relies on the fact that only this thread of + * control (running flushd_cloud) is manipulating the inode map, and so + * concurrent updates to the inode map are prevented even without the + * filesystem lock held. Take great care if allowing multi-threaded access to + * the inode map... */ +static void flushd_cloud(BlueSkyFS *fs) +{ + g_mutex_lock(fs->lock); + + /* TODO: Locking? Since we're reading a single variable this is probably + * atomic but a lock could be safer. */ + BlueSkyCloudLog *marker = bluesky_log_get_commit_point(fs); + int journal_seq_start = fs->log->seq_num; + + while (1) { + BlueSkyInode *inode; + if (fs->dirty_list.prev == NULL) + break; + inode = fs->dirty_list.prev->data; + + if (bluesky_verbose) { + g_log("bluesky/flushd", G_LOG_LEVEL_DEBUG, + "Flushing inode %"PRIu64" to cloud", inode->inum); + } + + bluesky_inode_ref(inode); + + g_mutex_unlock(fs->lock); + + g_mutex_lock(inode->lock); + g_assert(inode->change_cloud == inode->change_commit); + g_mutex_lock(fs->lock); + bluesky_list_unlink(&fs->dirty_list, inode->dirty_list); + inode->dirty_list = NULL; + g_mutex_unlock(fs->lock); + + BlueSkyCloudLog *log = inode->committed_item; + inode->committed_item = NULL; + g_mutex_unlock(inode->lock); + + if (log != NULL) + bluesky_cloudlog_serialize(log, fs); + bluesky_inode_unref(inode); + bluesky_cloudlog_unref(log); + + g_mutex_lock(fs->lock); + } + g_mutex_unlock(fs->lock); + + /* Write out any updated inode map entries, so that all inodes just written + * can be located, and then a final commit record. */ + BlueSkyCloudLog *commit_record = bluesky_inode_map_serialize(fs); + if (commit_record != NULL) { + bluesky_cloudlog_serialize(commit_record, fs); + } else { + g_print("No need for a checkpoint record...\n"); + } + + bluesky_cloudlog_flush(fs); + + /* Wait until all segments have been written to the cloud, so that it + * becomes safe to free up journal segments. */ + while (fs->log_state->pending_segments != NULL) { + SerializedRecord *segment + = (SerializedRecord *)fs->log_state->pending_segments->data; + g_mutex_lock(segment->lock); + while (!segment->complete) + g_cond_wait(segment->cond, segment->lock); + g_mutex_unlock(segment->lock); + + g_mutex_free(segment->lock); + g_cond_free(segment->cond); + g_free(segment); + + fs->log_state->pending_segments + = g_list_delete_link(fs->log_state->pending_segments, + fs->log_state->pending_segments); + } + + bluesky_log_write_commit_point(fs, marker); + bluesky_cloudlog_unref(commit_record); + + g_print("All segments have been flushed, journal < %d is clean\n", + journal_seq_start); + + fs->log->journal_watermark = journal_seq_start; + + bluesky_inode_map_minimize(fs); +} + /* Drop cached data for a given inode, if it is clean. inode must be locked. */ static void drop_caches(BlueSkyInode *inode) { if (inode->type == BLUESKY_REGULAR) bluesky_file_drop_cached(inode); + + BlueSkyCloudLog *log = inode->committed_item; + if (log != NULL) { + g_mutex_lock(log->lock); + if (log->data != NULL + && g_atomic_int_get(&log->data_lock_count) == 0 + && (log->location_flags != 0)) + { + bluesky_cloudlog_stats_update(log, -1); + bluesky_string_unref(log->data); + log->data = NULL; + bluesky_cloudlog_stats_update(log, 1); + } + g_mutex_unlock(log->lock); + } } -/* Drop clean data from the cache if needed due to memory pressure. */ -static void flushd_clean(BlueSkyFS *fs) +/* Drop clean data from the cache if needed. Clean data should generally be + * memory-mapped from log file or similar, so the kernel can drop this clean + * data from memory for us and hence memory management isn't too important. + * Mainly, we'll want to drop references to data that hasn't been accessed in a + * while so that it is possible to reclaim log segments on disk. + * + * If aggressive is set, try much harder to drop data from the caches to free + * up space. */ +static void flushd_clean(BlueSkyFS *fs, int aggressive) { g_mutex_lock(fs->lock); @@ -130,16 +248,15 @@ static void flushd_clean(BlueSkyFS *fs) inode_count = 1; while (inode_count-- > 0) { -#if 0 - if (g_atomic_int_get(&fs->cache_total) < bluesky_watermark_medium_total) - break; -#endif - BlueSkyInode *inode; if (fs->accessed_list.prev == NULL) break; inode = fs->accessed_list.prev->data; + uint64_t elapsed = bluesky_get_current_time() - inode->access_time; + if (elapsed < CACHE_DROP_DELAY && !aggressive) + break; + if (bluesky_verbose) { g_log("bluesky/flushd", G_LOG_LEVEL_DEBUG, "Considering dropping cached data for inode %"PRIu64, @@ -168,15 +285,132 @@ static void flushd_clean(BlueSkyFS *fs) g_mutex_unlock(fs->lock); } +/* Scan through all currently-stored files in the journal/cache and garbage + * collect old unused ones, if needed. */ +static void gather_cachefiles(gpointer key, gpointer value, gpointer user_data) +{ + GList **files = (GList **)user_data; + *files = g_list_prepend(*files, value); +} + +static gint compare_cachefiles(gconstpointer a, gconstpointer b) +{ + int64_t ta, tb; + + ta = ((BlueSkyCacheFile *)a)->atime; + tb = ((BlueSkyCacheFile *)b)->atime; + if (ta < tb) + return -1; + else if (ta > tb) + return 1; + else + return 0; +} + +void bluesky_cachefile_gc(BlueSkyFS *fs) +{ + GList *files = NULL; + + g_mutex_lock(fs->log->mmap_lock); + g_hash_table_foreach(fs->log->mmap_cache, gather_cachefiles, &files); + + /* Sort based on atime. The atime should be stable since it shouln't be + * updated except by threads which can grab the mmap_lock, which we already + * hold. */ + files = g_list_sort(files, compare_cachefiles); + + /* Walk the list of files, starting with the oldest, deleting files if + * possible until enough space has been reclaimed. */ + g_print("\nScanning cache: (total size = %d kB)\n", fs->log->disk_used); + while (files != NULL) { + BlueSkyCacheFile *cachefile = (BlueSkyCacheFile *)files->data; + /* Try to lock the structure, but if the lock is held by another thread + * then we'll just skip the file on this pass. */ + if (g_mutex_trylock(cachefile->lock)) { + int64_t age = bluesky_get_current_time() - cachefile->atime; + if (bluesky_verbose) { + g_print("%s addr=%p mapcount=%d refcount=%d size=%d atime_age=%f", + cachefile->filename, cachefile->addr, cachefile->mapcount, + cachefile->refcount, cachefile->disk_used, age / 1e6); + if (cachefile->fetching) + g_print(" (fetching)"); + g_print("\n"); + } + + gboolean deletion_candidate = FALSE; + if (g_atomic_int_get(&fs->log->disk_used) + > bluesky_options.cache_size + && g_atomic_int_get(&cachefile->refcount) == 0 + && g_atomic_int_get(&cachefile->mapcount) == 0) + { + deletion_candidate = TRUE; + } + + /* Don't allow journal files to be reclaimed until all data is + * known to be durably stored in the cloud. */ + if (cachefile->type == CLOUDLOG_JOURNAL + && cachefile->log_seq >= fs->log->journal_watermark) + { + deletion_candidate = FALSE; + } + + if (deletion_candidate) { + if (bluesky_verbose) { + g_print(" ...deleting\n"); + } + if (unlinkat(fs->log->dirfd, cachefile->filename, 0) < 0) { + fprintf(stderr, "Unable to unlink journal %s: %m\n", + cachefile->filename); + } + + g_atomic_int_add(&fs->log->disk_used, -cachefile->disk_used); + g_hash_table_remove(fs->log->mmap_cache, cachefile->filename); + bluesky_rangeset_free(cachefile->items); + if (cachefile->prefetches != NULL) + bluesky_rangeset_free(cachefile->prefetches); + g_mutex_unlock(cachefile->lock); + g_mutex_free(cachefile->lock); + g_cond_free(cachefile->cond); + g_free(cachefile->filename); + g_free(cachefile); + } else { + g_mutex_unlock(cachefile->lock); + } + } + files = g_list_delete_link(files, files); + } + g_list_free(files); + g_print("\nEnding cache size: %d kB\n", fs->log->disk_used); + + g_mutex_unlock(fs->log->mmap_lock); +} + /* Run the flush daemon for a single iteration, though if it is already * executing returns immediately. */ static gpointer flushd_task(BlueSkyFS *fs) { if (!g_mutex_trylock(fs->flushd_lock)) return NULL; + + g_print("\nCloudlog cache: %d dirty, %d writeback, %d journal, %d cloud\n", + g_atomic_int_get(&fs->cache_log_dirty), + g_atomic_int_get(&fs->cache_log_writeback), + g_atomic_int_get(&fs->cache_log_journal), + g_atomic_int_get(&fs->cache_log_cloud)); + flushd_dirty(fs); - bluesky_cloudlog_write_log(fs); - flushd_clean(fs); + flushd_cloud(fs); + flushd_clean(fs, 0); + bluesky_cachefile_gc(fs); + + /* If running out of disk cache space, make another more aggressive pass to + * free up space. */ + if (g_atomic_int_get(&fs->log->disk_used) > bluesky_options.cache_size) { + g_print("Still short on disk space, trying again to free space...\n"); + flushd_clean(fs, 1); + bluesky_cachefile_gc(fs); + } + g_mutex_unlock(fs->flushd_lock); return NULL; @@ -187,20 +421,101 @@ void bluesky_flushd_invoke(BlueSkyFS *fs) g_thread_create((GThreadFunc)flushd_task, fs, FALSE, NULL); } +/* How urgent is flushing out data? Returns one of several values: + * 0 - memory state is fine + * 1 - should launch flushd if not already running + * 2 - should delay writers while memory frees up + * 3 - should block writers until memory frees up + */ +static int compute_pressure(BlueSkyFS *fs) +{ + /* LEVEL 3 */ + /* Too much dirty data in memory? */ + if (g_atomic_int_get(&fs->cache_dirty) + + g_atomic_int_get(&fs->cache_log_dirty) + > bluesky_watermark_high_dirty) + { + g_print("pressure: too much dirty data (3)\n"); + return 3; + } + + /* Too much uncommitted data in the journal on disk, not yet flushed to the + * cloud? */ + /*printf("Dirty journals: %d to %d\n", + fs->log->journal_watermark, fs->log->seq_num);*/ + int dirty_limit; + dirty_limit = bluesky_options.cache_size / (LOG_SEGMENT_SIZE / 1024) / 2; + int dirty_journals = fs->log->seq_num - fs->log->journal_watermark + 1; + if (dirty_journals > 1 && dirty_journals >= dirty_limit) { + printf("pressure: too many dirty journals (%d >= %d) (3)\n", + dirty_journals, dirty_limit); + return 3; + } + + /* LEVEL 2 */ + if (g_atomic_int_get(&fs->cache_dirty) > bluesky_watermark_med2_dirty) { + g_print("pressure: too much dirty data (2)\n"); + return 2; + } + + if (dirty_journals > 1 && dirty_journals > dirty_limit * 3 / 4) { + printf("pressure: many dirty journals (%d), should start writeback (2)\n", + dirty_journals); + return 2; + } + + /* LEVEL 1 */ + if (g_atomic_int_get(&fs->cache_dirty) > bluesky_watermark_medium_dirty) { + g_print("pressure: too much dirty data (1)\n"); + return 1; + } + + if (dirty_journals > 1 && dirty_journals > dirty_limit / 2) { + printf("pressure: many dirty journals (%d), should start writeback (1)\n", + dirty_journals); + return 1; + } + + return 0; +} + void bluesky_flushd_invoke_conditional(BlueSkyFS *fs) { - if (g_atomic_int_get(&fs->cache_dirty) < bluesky_watermark_high_dirty - && g_atomic_int_get(&fs->cache_total) < bluesky_watermark_high_total) + int pressure = compute_pressure(fs); + if (pressure == 0) return; if (bluesky_verbose) { g_log("bluesky/flushd", G_LOG_LEVEL_DEBUG, - "Too much data; invoking flushd: dirty=%d total=%d", - g_atomic_int_get(&fs->cache_dirty), - g_atomic_int_get(&fs->cache_total)); + "Too much data; invoking flushd: dirty=%d", + g_atomic_int_get(&fs->cache_dirty)); } bluesky_flushd_invoke(fs); + + /* If the system is under heavy memory pressure, actually delay execution + * so the flush daemon can catch up. */ + while (pressure > 1) { + g_log("bluesky/flushd", G_LOG_LEVEL_DEBUG, + "Waiting due to memory pressure, dirty=%d + %d", + g_atomic_int_get(&fs->cache_dirty), + g_atomic_int_get(&fs->cache_log_dirty)); + g_mutex_lock(fs->lock); + pressure = compute_pressure(fs); + if (pressure > 2) { + g_cond_wait(fs->flushd_cond, fs->lock); + } else if (pressure > 1) { + /* Wait for up to 10 seconds. */ + GTimeVal timeout; + g_get_current_time(&timeout); + g_time_val_add(&timeout, 10 * 1000 * 1000); + g_cond_timed_wait(fs->flushd_cond, fs->lock, &timeout); + } + g_mutex_unlock(fs->lock); + pressure = compute_pressure(fs); + if (pressure == 1) + break; /* Do not loop indefinitely for a pressure of 1 */ + } } /* Start a perpetually-running thread that flushes the cache occasionally. */