Fix Order Bug (#301)

* fix order bug

* beta

* more logging, new beta

* fix version

* fix xl order call

* add index to rid column

* version

CHANGELOG.rst

@@ -6,6 +6,13 @@ snovault
 Change Log
 ----------
 
+11.20.0
+=======
+
+* Bug fix: use loadxl_order() in staggered reindexing
+* Add B-tree index to rid column in propsheets to optimize revision history retrieval
+
+
 11.19.0
 =======
 

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "dcicsnovault"
-version = "11.19.0"
+version = "11.20.0"
 description = "Storage support for 4DN Data Portals."
 authors = ["4DN-DCIC Team <[email protected]>"]
 license = "MIT"

snovault/commands/create_mapping_on_deploy.py

@@ -66,7 +66,7 @@ def _run_create_mapping(app, args):
             log.info('Overriding deploy_cfg and wiping ES')
             deploy_cfg['WIPE_ES'] = True
         run_create_mapping(app, check_first=(not deploy_cfg['WIPE_ES']), purge_queue=args.clear_queue,
-                           item_order=loadxl_order)
+                           item_order=loadxl_order())
     except Exception as e:
         log.error("Exception encountered while gathering deployment information or running create_mapping")
         log.error(str(e))

snovault/dev_servers.py

@@ -213,6 +213,8 @@ def cleanup_process():
 
         # now clear the queues and queue items for indexing
         create_mapping.run(app, check_first=True, strict=True, purge_queue=False)
+        # To test prod setup:
+        # create_mapping.reindex_by_type_staggered(app)
 
     PRINT('Started. ^C to exit.')
 

snovault/elasticsearch/create_mapping.py

@@ -33,6 +33,7 @@
 # from ..commands.es_index_data import run as run_index_data
 from ..schema_utils import combine_schemas
 from ..tools import make_indexer_testapp
+from ..project_app import app_project
 from ..util import (
     add_default_embeds, IndexSettings,
     NUM_SHARDS, NUM_REPLICAS, SEARCH_MAX, KW_IGNORE_ABOVE, MIN_NGRAM,
@@ -874,11 +875,11 @@ def build_index(app, es, index_name, in_type, mapping, uuids_to_index, dry_run,
     start = timer()
     coll_uuids = set(get_uuids_for_types(app.registry, types=[in_type]))
     end = timer()
-    log.info('Time to get collection uuids: %s' % str(end-start), cat='fetch time',
-             duration=str(end-start), collection=in_type)
+    log.warning('Time to get collection uuids: %s' % str(end-start), cat='fetch time',
+                duration=str(end-start), collection=in_type)
     uuids_to_index[in_type] = coll_uuids
-    log.info('MAPPING: will queue all %s items in the new index %s for reindexing' %
-             (len(coll_uuids), in_type), cat='items to queue', count=len(coll_uuids), collection=in_type)
+    log.warning('MAPPING: will queue all %s items in the new index %s for reindexing' %
+                (len(coll_uuids), in_type), cat='items to queue', count=len(coll_uuids), collection=in_type)
 
 
 def check_if_index_exists(es, in_type):
@@ -1383,7 +1384,7 @@ def reindex_by_type_staggered(app):
     registry = app.registry
     es = registry[ELASTIC_SEARCH]
     indexer_queue = registry[INDEXER_QUEUE]
-    all_types = registry[COLLECTIONS].by_item_type
+    all_types = app_project().loadxl_order()
 
     log.warning('Running staggered create_mapping command - wiping and reindexing indices sequentially'
                 ' to minimize downtime.')
@@ -1396,10 +1397,11 @@ def reindex_by_type_staggered(app):
         uuids = {}
         build_index(app, es, namespaced_index, i_type, mapping, uuids, False)
         mapping_end = timer()
-        to_index_list = flatten_and_sort_uuids(app.registry, uuids, None)
+        to_index_list = flatten_and_sort_uuids(app.registry, uuids, None)  # none here is fine since we pre-ordered
         indexer_queue.add_uuids(app.registry, to_index_list, strict=True,
                                 target_queue='secondary')
-        log.warning(f'Queued type {i_type} in {mapping_end - current_start}')
+        log.warning(f'Queued type {i_type} ({len(to_index_list)} total items) in {mapping_end - current_start}')
+        log.warning(f'First 10 items: {list(uuid for uuid in to_index_list[0:10])}')
         time.sleep(10)  # give queue some time to catch up
         while not indexer_queue.queue_is_empty():
             time.sleep(10)  # check every 10 seconds

snovault/storage.py

@@ -784,11 +784,12 @@ class PropertySheet(Base):
     )
     # The sid column also serves as the order.
     sid = Column(types.Integer, autoincrement=True, primary_key=True)
+    # B-tree index on rid here greatly optimizes retrieval of revision history
     rid = Column(UUID,
                  ForeignKey('resources.rid',
                             deferrable=True,
                             initially='DEFERRED'),
-                 nullable=False)
+                 nullable=False, index=True)
     name = Column(types.String, nullable=False)
     properties = Column(JSON)
     resource = orm.relationship('Resource')