First attempt at supporting reading data back from cloud log segments.
[bluesky.git] / bluesky / cloudlog.c
index 76776b4..0c3fb02 100644 (file)
@@ -93,6 +93,25 @@ BlueSkyCloudLog *bluesky_cloudlog_new(BlueSkyFS *fs)
     return log;
 }
 
+/* Helper function for updating memory usage statistics for a filesystem (the
+ * cache_log_* variables).  This will increment (type=1) or decrement (type=-1)
+ * the counter associated with the current state of the cloud log item.  The
+ * item should be locked or otherwise protected from concurrent access. */
+void bluesky_cloudlog_stats_update(BlueSkyCloudLog *log, int type)
+{
+    BlueSkyFS *fs = log->fs;
+
+    if (log->location_flags & CLOUDLOG_CLOUD) {
+        g_atomic_int_add(&fs->cache_log_cloud, type);
+    } else if (log->location_flags & CLOUDLOG_JOURNAL) {
+        g_atomic_int_add(&fs->cache_log_journal, type);
+    } else if (log->pending_write & CLOUDLOG_JOURNAL) {
+        g_atomic_int_add(&fs->cache_log_journal, type);
+    } else if (log->data != NULL) {
+        g_atomic_int_add(&fs->cache_log_dirty, type);
+    }
+}
+
 /* The reference held by the hash table does not count towards the reference
  * count.  When a new object is created, it initially has a reference count of
  * 1 for the creator, and similarly fetching an item from the hash table will
@@ -128,6 +147,7 @@ void bluesky_cloudlog_unref(BlueSkyCloudLog *log)
         g_hash_table_remove(fs->locations, &log->id);
         g_mutex_unlock(fs->lock);
 
+        bluesky_cloudlog_stats_update(log, -1);
         log->type = LOGTYPE_INVALID;
         g_mutex_free(log->lock);
         g_cond_free(log->cond);
@@ -182,10 +202,21 @@ void bluesky_cloudlog_fetch(BlueSkyCloudLog *log)
     if (log->data != NULL)
         return;
 
-    g_assert(log->location_flags & CLOUDLOG_JOURNAL);
-
-    log->data = bluesky_log_map_object(log->fs->log, log->log_seq,
-                                       log->log_offset, log->log_size);
+    if ((log->location_flags | log->pending_write) & CLOUDLOG_JOURNAL) {
+        bluesky_cloudlog_stats_update(log, -1);
+        log->data = bluesky_log_map_object(log->fs, -1, log->log_seq,
+                                           log->log_offset, log->log_size);
+        bluesky_cloudlog_stats_update(log, 1);
+    } else if (log->location_flags & CLOUDLOG_CLOUD) {
+        bluesky_cloudlog_stats_update(log, -1);
+        log->data = bluesky_log_map_object(log->fs, log->location.directory,
+                                           log->location.sequence,
+                                           log->location.offset,
+                                           log->location.size);
+        bluesky_cloudlog_stats_update(log, 1);
+    } else {
+        g_error("Unable to fetch cloudlog entry!");
+    }
 
     g_cond_broadcast(log->cond);
 }
@@ -210,11 +241,15 @@ BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
     bluesky_cloudlog_fetch(log);
     g_assert(log->data != NULL);
 
+    bluesky_cloudlog_stats_update(log, -1);
+
+    /* TODO: Right now offset/size are set to the raw data, but we should add
+     * header parsing to the code which loads objects back in. */
     log->location = state->location;
-    log->location.offset = state->data->len;
-    log->location.size
-        = sizeof(struct log_header) + sizeof(BlueSkyCloudID) * 0
-           + log->data->len;
+    log->location.offset = state->data->len + sizeof(struct log_header);
+    log->location.size = log->data->len;
+        /* = sizeof(struct log_header) + sizeof(BlueSkyCloudID) * 0
+           + log->data->len; */
 
     struct log_header header;
     memcpy(header.magic, "AgI ", 4);
@@ -226,6 +261,7 @@ BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
     g_string_append_len(state->data, log->data->data, log->data->len);
 
     log->location_flags |= CLOUDLOG_CLOUD;
+    bluesky_cloudlog_stats_update(log, 1);
     g_mutex_unlock(log->lock);
 
     if (state->data->len > CLOUDLOG_SEGMENT_SIZE)