Blob Blame Raw
From 8355b844dbc7097ddc5639a1da0932c60ca50aee Mon Sep 17 00:00:00 2001
From: Mark Reynolds <mreynolds@redhat.com>
Date: Wed, 16 Oct 2019 20:27:30 -0400
Subject: [PATCH] Issue 49850 -  ldbm_get_nonleaf_ids() slow for databases with
 many non-leaf entries

Bug Description:  The logs from an LDIF import indicated that gathering non-leaf IDs
                  for creating the ancestorid index took an enormous amount of time,
                  over 10hrs.  The root cause is that the parentid index btree ordering
                  is lexical, but the IDList being built up from it is sorted numerically.
                  In the existing code, the IDList is maintained in constantly sorted
                  order by idl_insert().

Fix Description:  ldbm_get_nonleaf_ids() switches to idl_append_extend() instead idl_insert()
                  for building up the IDList and then sorts the result only once, using
                  qsort with idl_sort_cmp, after the entire list has been gathered.

                  The improvement on identical hardware is for the operation to take 10
                  seconds rather than 10 hours

Patch Author:  Thomas Lackey <telackey@bozemanpass.com>  Thanks for the great contribution!!!

relates: https://pagure.io/389-ds-base/issue/49850

Reviewed by: mreynolds, tbordaz, and firstyear
---
 ldap/servers/slapd/back-ldbm/ancestorid.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

diff --git a/ldap/servers/slapd/back-ldbm/ancestorid.c b/ldap/servers/slapd/back-ldbm/ancestorid.c
index 24642923d..254a3aa3b 100644
--- a/ldap/servers/slapd/back-ldbm/ancestorid.c
+++ b/ldap/servers/slapd/back-ldbm/ancestorid.c
@@ -82,7 +82,14 @@ ldbm_get_nonleaf_ids(backend *be, DB_TXN *txn, IDList **idl, ImportJob *job)
         ret = dbc->c_get(dbc, &key, &data, DB_NEXT_NODUP);
         if ((ret == 0) && (*(char *)key.data == EQ_PREFIX)) {
             id = (ID)strtoul((char *)key.data + 1, NULL, 10);
-            idl_insert(&nodes, id);
+            /*
+             * TEL 20180711 - switch to idl_append instead of idl_insert because there is no
+             * no need to keep the list constantly sorted, which can be very expensive with
+             * large databases (exacerbated by the fact that the parentid btree ordering is
+             * lexical, but the idl_insert ordering is numeric).  It is enough to gather them
+             * all together and sort them once at the end.
+             */
+            idl_append_extend(&nodes, id);
         }
         key_count++;
         if (!(key_count % PROGRESS_INTERVAL)) {
@@ -107,6 +114,17 @@ ldbm_get_nonleaf_ids(backend *be, DB_TXN *txn, IDList **idl, ImportJob *job)
     if (ret != 0)
         ldbm_nasty("ldbm_get_nonleaf_ids", sourcefile, 13030, ret);
 
+    if (ret == 0) {
+        /* now sort it */
+        import_log_notice(job, SLAPI_LOG_INFO, "ldbm_get_nonleaf_ids",
+            "Starting sort of ancestorid non-leaf IDs...");
+
+        qsort((void *)&nodes->b_ids[0], nodes->b_nids, (size_t)sizeof(ID), idl_sort_cmp);
+
+        import_log_notice(job, SLAPI_LOG_INFO, "ldbm_get_nonleaf_ids",
+            "Finished sort of ancestorid non-leaf IDs.");
+    }
+
 out:
     /* Close the cursor */
     if (dbc != NULL) {
-- 
2.21.0