389-ds-base/0001-Issue-7096-During-replication-online-total-init-the-.patch
Viktor Ashirov 6a9fe72ef8 Fix broken FreeIPA upgrade and replication issues
- Resolves: rhbz#2424132 Upgrade from freeipa-4.12.5-3 and 390-ds-base-3.1.3-10 to latest rawhide fails
- Fix jemalloc compilation issue with GCC 15
- Issue 7096 - During replication online total init the function idl_id_is_in_idlist is not scaling with large database
- Issue 7118 - Revise paged result search locking
- Issue 7108 - Fix shutdown crash in entry cache destruction

Use correct tarball from upstream.
2026-01-09 18:48:11 +01:00

318 lines
11 KiB
Diff

From 1c9c535888b9a850095794787d67900b04924a76 Mon Sep 17 00:00:00 2001
From: tbordaz <tbordaz@redhat.com>
Date: Wed, 7 Jan 2026 11:21:12 +0100
Subject: [PATCH] Issue 7096 - During replication online total init the
function idl_id_is_in_idlist is not scaling with large database (#7145)
Bug description:
During a online total initialization, the supplier sorts
the candidate list of entries so that the parents are sent before
children entries.
With large DB the ID array used for the sorting is not
scaling. It takes so long to build the candidate list that
the connection gets closed
Fix description:
Instead of using an ID array, uses a list of ID ranges
fixes: #7096
Reviewed by: Mark Reynolds, Pierre Rogier (Thanks !!)
---
ldap/servers/slapd/back-ldbm/back-ldbm.h | 12 ++
ldap/servers/slapd/back-ldbm/idl_common.c | 163 ++++++++++++++++++
ldap/servers/slapd/back-ldbm/idl_new.c | 30 ++--
.../servers/slapd/back-ldbm/proto-back-ldbm.h | 3 +
4 files changed, 189 insertions(+), 19 deletions(-)
diff --git a/ldap/servers/slapd/back-ldbm/back-ldbm.h b/ldap/servers/slapd/back-ldbm/back-ldbm.h
index 1bc36720d..b187c26bc 100644
--- a/ldap/servers/slapd/back-ldbm/back-ldbm.h
+++ b/ldap/servers/slapd/back-ldbm/back-ldbm.h
@@ -282,6 +282,18 @@ typedef struct _idlist_set
#define INDIRECT_BLOCK(idl) ((idl)->b_nids == INDBLOCK)
#define IDL_NIDS(idl) (idl ? (idl)->b_nids : (NIDS)0)
+/*
+ * used by the supplier during online total init
+ * it stores the ranges of ID that are already present
+ * in the candidate list ('parentid>=1')
+ */
+typedef struct IdRange {
+ ID first;
+ ID last;
+ struct IdRange *next;
+} IdRange_t;
+
+
typedef size_t idl_iterator;
/* small hashtable implementation used in the entry cache -- the table
diff --git a/ldap/servers/slapd/back-ldbm/idl_common.c b/ldap/servers/slapd/back-ldbm/idl_common.c
index fcb0ece4b..fdc9b4e67 100644
--- a/ldap/servers/slapd/back-ldbm/idl_common.c
+++ b/ldap/servers/slapd/back-ldbm/idl_common.c
@@ -172,6 +172,169 @@ idl_min(IDList *a, IDList *b)
return (a->b_nids > b->b_nids ? b : a);
}
+/*
+ * This is a faster version of idl_id_is_in_idlist.
+ * idl_id_is_in_idlist uses an array of ID so lookup is expensive
+ * idl_id_is_in_idlist_ranges uses a list of ranges of ID lookup is faster
+ * returns
+ * 1: 'id' is present in idrange_list
+ * 0: 'id' is not present in idrange_list
+ */
+int
+idl_id_is_in_idlist_ranges(IDList *idl, IdRange_t *idrange_list, ID id)
+{
+ IdRange_t *range = idrange_list;
+ int found = 0;
+
+ if (NULL == idl || NOID == id) {
+ return 0; /* not in the list */
+ }
+ if (ALLIDS(idl)) {
+ return 1; /* in the list */
+ }
+
+ for(;range; range = range->next) {
+ if (id > range->last) {
+ /* check if it belongs to the next range */
+ continue;
+ }
+ if (id >= range->first) {
+ /* It belongs to that range [first..last ] */
+ found = 1;
+ break;
+ } else {
+ /* this range is after id */
+ break;
+ }
+ }
+ return found;
+}
+
+/* This function is used during the online total initialisation
+ * (see next function)
+ * It frees all ranges of ID in the list
+ */
+void idrange_free(IdRange_t **head)
+{
+ IdRange_t *curr, *sav;
+
+ if ((head == NULL) || (*head == NULL)) {
+ return;
+ }
+ curr = *head;
+ sav = NULL;
+ for (; curr;) {
+ sav = curr;
+ curr = curr->next;
+ slapi_ch_free((void *) &sav);
+ }
+ if (sav) {
+ slapi_ch_free((void *) &sav);
+ }
+ *head = NULL;
+}
+
+/* This function is used during the online total initialisation
+ * Because a MODRDN can move entries under a parent that
+ * has a higher ID we need to sort the IDList so that parents
+ * are sent, to the consumer, before the children are sent.
+ * The sorting with a simple IDlist does not scale instead
+ * a list of IDs ranges is much faster.
+ * In that list we only ADD/lookup ID.
+ */
+IdRange_t *idrange_add_id(IdRange_t **head, ID id)
+{
+ if (head == NULL) {
+ slapi_log_err(SLAPI_LOG_ERR, "idrange_add_id",
+ "Can not add ID %d in non defined list\n", id);
+ return NULL;
+ }
+
+ if (*head == NULL) {
+ /* This is the first range */
+ IdRange_t *new_range = (IdRange_t *)slapi_ch_malloc(sizeof(IdRange_t));
+ new_range->first = id;
+ new_range->last = id;
+ new_range->next = NULL;
+ *head = new_range;
+ return *head;
+ }
+
+ IdRange_t *curr = *head, *prev = NULL;
+
+ /* First, find if id already falls within any existing range, or it is adjacent to any */
+ while (curr) {
+ if (id >= curr->first && id <= curr->last) {
+ /* inside a range, nothing to do */
+ return curr;
+ }
+
+ if (id == curr->last + 1) {
+ /* Extend this range upwards */
+ curr->last = id;
+
+ /* Check for possible merge with next range */
+ IdRange_t *next = curr->next;
+ if (next && curr->last + 1 >= next->first) {
+ slapi_log_err(SLAPI_LOG_REPL, "idrange_add_id",
+ "(id=%d) merge current with next range [%d..%d]\n", id, curr->first, curr->last);
+ curr->last = (next->last > curr->last) ? next->last : curr->last;
+ curr->next = next->next;
+ slapi_ch_free((void*) &next);
+ } else {
+ slapi_log_err(SLAPI_LOG_REPL, "idrange_add_id",
+ "(id=%d) extend forward current range [%d..%d]\n", id, curr->first, curr->last);
+ }
+ return curr;
+ }
+
+ if (id + 1 == curr->first) {
+ /* Extend this range downwards */
+ curr->first = id;
+
+ /* Check for possible merge with previous range */
+ if (prev && prev->last + 1 >= curr->first) {
+ prev->last = curr->last;
+ prev->next = curr->next;
+ slapi_ch_free((void *) &curr);
+ slapi_log_err(SLAPI_LOG_REPL, "idrange_add_id",
+ "(id=%d) merge current with previous range [%d..%d]\n", id, prev->first, prev->last);
+ return prev;
+ } else {
+ slapi_log_err(SLAPI_LOG_REPL, "idrange_add_id",
+ "(id=%d) extend backward current range [%d..%d]\n", id, curr->first, curr->last);
+ return curr;
+ }
+ }
+
+ /* If id is before the current range, break so we can insert before */
+ if (id < curr->first) {
+ break;
+ }
+
+ prev = curr;
+ curr = curr->next;
+ }
+ /* Need to insert a new standalone IdRange */
+ IdRange_t *new_range = (IdRange_t *)slapi_ch_malloc(sizeof(IdRange_t));
+ new_range->first = id;
+ new_range->last = id;
+ new_range->next = curr;
+
+ if (prev) {
+ slapi_log_err(SLAPI_LOG_REPL, "idrange_add_id",
+ "(id=%d) add new range [%d..%d]\n", id, new_range->first, new_range->last);
+ prev->next = new_range;
+ } else {
+ /* Insert at head */
+ slapi_log_err(SLAPI_LOG_REPL, "idrange_add_id",
+ "(id=%d) head range [%d..%d]\n", id, new_range->first, new_range->last);
+ *head = new_range;
+ }
+ return *head;
+}
+
+
int
idl_id_is_in_idlist(IDList *idl, ID id)
{
diff --git a/ldap/servers/slapd/back-ldbm/idl_new.c b/ldap/servers/slapd/back-ldbm/idl_new.c
index 5fbcaff2e..2d978353f 100644
--- a/ldap/servers/slapd/back-ldbm/idl_new.c
+++ b/ldap/servers/slapd/back-ldbm/idl_new.c
@@ -417,7 +417,6 @@ idl_new_range_fetch(
{
int ret = 0;
int ret2 = 0;
- int idl_rc = 0;
dbi_cursor_t cursor = {0};
IDList *idl = NULL;
dbi_val_t cur_key = {0};
@@ -436,6 +435,7 @@ idl_new_range_fetch(
size_t leftoverlen = 32;
size_t leftovercnt = 0;
char *index_id = get_index_name(be, db, ai);
+ IdRange_t *idrange_list = NULL;
if (NULL == flag_err) {
@@ -578,10 +578,12 @@ idl_new_range_fetch(
* found entry is the one from the suffix
*/
suffix = key;
- idl_rc = idl_append_extend(&idl, id);
- } else if ((key == suffix) || idl_id_is_in_idlist(idl, key)) {
+ idl_append_extend(&idl, id);
+ idrange_add_id(&idrange_list, id);
+ } else if ((key == suffix) || idl_id_is_in_idlist_ranges(idl, idrange_list, key)) {
/* the parent is the suffix or already in idl. */
- idl_rc = idl_append_extend(&idl, id);
+ idl_append_extend(&idl, id);
+ idrange_add_id(&idrange_list, id);
} else {
/* Otherwise, keep the {key,id} in leftover array */
if (!leftover) {
@@ -596,13 +598,7 @@ idl_new_range_fetch(
leftovercnt++;
}
} else {
- idl_rc = idl_append_extend(&idl, id);
- }
- if (idl_rc) {
- slapi_log_err(SLAPI_LOG_ERR, "idl_new_range_fetch",
- "Unable to extend id list (err=%d)\n", idl_rc);
- idl_free(&idl);
- goto error;
+ idl_append_extend(&idl, id);
}
count++;
@@ -695,21 +691,17 @@ error:
while(remaining > 0) {
for (size_t i = 0; i < leftovercnt; i++) {
- if (leftover[i].key > 0 && idl_id_is_in_idlist(idl, leftover[i].key) != 0) {
+ if (leftover[i].key > 0 && idl_id_is_in_idlist_ranges(idl, idrange_list, leftover[i].key) != 0) {
/* if the leftover key has its parent in the idl */
- idl_rc = idl_append_extend(&idl, leftover[i].id);
- if (idl_rc) {
- slapi_log_err(SLAPI_LOG_ERR, "idl_new_range_fetch",
- "Unable to extend id list (err=%d)\n", idl_rc);
- idl_free(&idl);
- return NULL;
- }
+ idl_append_extend(&idl, leftover[i].id);
+ idrange_add_id(&idrange_list, leftover[i].id);
leftover[i].key = 0;
remaining--;
}
}
}
slapi_ch_free((void **)&leftover);
+ idrange_free(&idrange_list);
}
slapi_log_err(SLAPI_LOG_FILTER, "idl_new_range_fetch",
"Found %d candidates; error code is: %d\n",
diff --git a/ldap/servers/slapd/back-ldbm/proto-back-ldbm.h b/ldap/servers/slapd/back-ldbm/proto-back-ldbm.h
index 91d61098a..30a7aa11f 100644
--- a/ldap/servers/slapd/back-ldbm/proto-back-ldbm.h
+++ b/ldap/servers/slapd/back-ldbm/proto-back-ldbm.h
@@ -217,6 +217,9 @@ ID idl_firstid(IDList *idl);
ID idl_nextid(IDList *idl, ID id);
int idl_init_private(backend *be, struct attrinfo *a);
int idl_release_private(struct attrinfo *a);
+IdRange_t *idrange_add_id(IdRange_t **head, ID id);
+void idrange_free(IdRange_t **head);
+int idl_id_is_in_idlist_ranges(IDList *idl, IdRange_t *idrange_list, ID id);
int idl_id_is_in_idlist(IDList *idl, ID id);
idl_iterator idl_iterator_init(const IDList *idl);
--
2.52.0