health check: use /<index>/_count endpoint; verify shards

In actual production verification, the /_mapping endpoint didn't seem to
work.

1 files changed, 12 insertions, 7 deletions

diff --git a/fatcat_scholar/search.py b/fatcat_scholar/search.py
index e9f2bc8..c52cae3 100644
--- a/fatcat_scholar/search.py
+++ b/fatcat_scholar/search.py
@@ -420,22 +420,27 @@ def do_fulltext_search(
 def es_scholar_index_alive() -> bool:
     """
     Checks if the configured back-end elasticsearch index exists and can
-    service queries. Intended to be used in health checks.
+    service queries. Intended to be used in health checks, called every couple
+    seconds.
 
     Note that the regular client.indices.exists(index) function call will
     return an error if the cluster leader can not be reached, even if the local
     node could service queries in a read-only manner.
 
-    The client.indices.get_mapping(index) API, or the client.cat.count(index)
-    API, both return quickly and indicate that queries can be run against the
-    index.
+    The client.count(body=None, index=index) API returns quickly enough (though
+    might be slow during indexing?), and returns context about the number of
+    shards queried, and thus seems like a good fit for this check.
     """
     try:
-        resp = es_client.indices.get_mapping(settings.ELASTICSEARCH_QUERY_FULLTEXT_INDEX)
+        resp = es_client.count(
+            body=None, index=settings.ELASTICSEARCH_QUERY_FULLTEXT_INDEX
+        )
     except elasticsearch.exceptions.RequestError as e_raw:
         if e_raw.status_code == 404:
             return False
         else:
             raise e_raw
-    return resp
-
+    try:
+        return bool(resp["_shards"]["successful"] == resp["_shards"]["total"])
+    except KeyError:
+        return False