Commit 234230f2 authored by Ondřej Kuzník's avatar Ondřej Kuzník Committed by Quanah Gibson-Mount
Browse files

ITS#8486 Switch sessionlog to use TAVL

parent a7fb531e
......@@ -117,7 +117,6 @@ typedef struct syncmatches {
/* Session log data */
typedef struct slog_entry {
struct slog_entry *se_next;
struct berval se_uuid;
struct berval se_csn;
int se_sid;
......@@ -131,9 +130,8 @@ typedef struct sessionlog {
int sl_num;
int sl_size;
int sl_playing;
slog_entry *sl_head;
slog_entry *sl_tail;
ldap_pvt_thread_mutex_t sl_mutex;
Avlnode *sl_entries;
ldap_pvt_thread_rdwr_t sl_mutex;
} sessionlog;
/* The main state for this overlay */
......@@ -402,6 +400,25 @@ sp_avl_cmp( const void *c1, const void *c2 )
return ber_bvcmp( &m1->mt_dn, &m2->mt_dn );
}
static int
syncprov_sessionlog_cmp( const void *l, const void *r )
{
const slog_entry *left = l, *right = r;
int ret = ber_bvcmp( &left->se_csn, &right->se_csn );
if ( !ret )
ret = ber_bvcmp( &left->se_uuid, &right->se_uuid );
/* Only time we have two modifications with same CSN is when we detect a
* rename during replication.
* We invert the test here because LDAP_REQ_MODDN is
* numerically greater than LDAP_REQ_MODIFY but we
* want it to occur first.
*/
if ( !ret )
ret = right->se_tag - left->se_tag;
return ret;
}
/* syncprov_findbase:
* finds the true DN of the base of a search (with alias dereferencing) and
* checks to make sure the base entry doesn't get replaced with a different
......@@ -1577,6 +1594,7 @@ syncprov_add_slog( Operation *op )
syncprov_info_t *si = on->on_bi.bi_private;
sessionlog *sl;
slog_entry *se;
int rc;
sl = si->si_logs;
{
......@@ -1586,24 +1604,20 @@ syncprov_add_slog( Operation *op )
* state with respect to such operations, so we ignore them and
* wipe out anything in the log if we see them.
*/
ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wlock( &sl->sl_mutex );
/* can only do this if no one else is reading the log at the moment */
if (!sl->sl_playing) {
while ( se = sl->sl_head ) {
sl->sl_head = se->se_next;
ch_free( se );
}
sl->sl_tail = NULL;
sl->sl_num = 0;
if ( !sl->sl_playing ) {
tavl_free( sl->sl_entries, (AVL_FREE)ch_free );
sl->sl_num = 0;
sl->sl_entries = NULL;
}
ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wunlock( &sl->sl_mutex );
return;
}
/* Allocate a record. UUIDs are not NUL-terminated. */
se = ch_malloc( sizeof( slog_entry ) + opc->suuid.bv_len +
se = ch_malloc( sizeof( slog_entry ) + opc->suuid.bv_len +
op->o_csn.bv_len + 1 );
se->se_next = NULL;
se->se_tag = op->o_tag;
se->se_uuid.bv_val = (char *)(&se[1]);
......@@ -1616,7 +1630,7 @@ syncprov_add_slog( Operation *op )
se->se_csn.bv_len = op->o_csn.bv_len;
se->se_sid = slap_parse_csn_sid( &se->se_csn );
ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wlock( &sl->sl_mutex );
if ( LogTest( LDAP_DEBUG_SYNC ) ) {
char uuidstr[40] = {};
if ( !BER_BVISEMPTY( &opc->suuid ) ) {
......@@ -1628,25 +1642,7 @@ syncprov_add_slog( Operation *op )
"adding csn=%s to sessionlog, uuid=%s\n",
op->o_log_prefix, se->se_csn.bv_val, uuidstr );
}
if ( sl->sl_head ) {
/* Keep the list in csn order. */
if ( ber_bvcmp( &sl->sl_tail->se_csn, &se->se_csn ) <= 0 ) {
sl->sl_tail->se_next = se;
sl->sl_tail = se;
} else {
slog_entry **sep;
for ( sep = &sl->sl_head; *sep; sep = &(*sep)->se_next ) {
if ( ber_bvcmp( &se->se_csn, &(*sep)->se_csn ) < 0 ) {
se->se_next = *sep;
*sep = se;
break;
}
}
}
} else {
sl->sl_head = se;
sl->sl_tail = se;
if ( !sl->sl_entries ) {
if ( !sl->sl_mincsn ) {
sl->sl_numcsns = 1;
sl->sl_mincsn = ch_malloc( 2*sizeof( struct berval ));
......@@ -1656,35 +1652,40 @@ syncprov_add_slog( Operation *op )
BER_BVZERO( &sl->sl_mincsn[1] );
}
}
rc = tavl_insert( &sl->sl_entries, se, syncprov_sessionlog_cmp, avl_dup_error );
assert( rc == LDAP_SUCCESS );
sl->sl_num++;
if (!sl->sl_playing) {
while ( sl->sl_num > sl->sl_size ) {
int i;
se = sl->sl_head;
sl->sl_head = se->se_next;
Debug( LDAP_DEBUG_SYNC, "%s syncprov_add_slog: "
"expiring csn=%s from sessionlog (sessionlog size=%d)\n",
op->o_log_prefix, se->se_csn.bv_val, sl->sl_num );
for ( i=0; i<sl->sl_numcsns; i++ )
if ( sl->sl_sids[i] >= se->se_sid )
break;
if ( i == sl->sl_numcsns || sl->sl_sids[i] != se->se_sid ) {
if ( !sl->sl_playing && sl->sl_num > sl->sl_size ) {
Avlnode *edge = tavl_end( sl->sl_entries, TAVL_DIR_LEFT );
while ( sl->sl_num > sl->sl_size ) {
int i;
Avlnode *next = tavl_next( edge, TAVL_DIR_RIGHT );
se = edge->avl_data;
Debug( LDAP_DEBUG_SYNC, "%s syncprov_add_slog: "
"adding csn=%s to mincsn\n",
op->o_log_prefix, se->se_csn.bv_val, 0 );
slap_insert_csn_sids( (struct sync_cookie *)sl,
i, se->se_sid, &se->se_csn );
} else {
Log4( LDAP_DEBUG_SYNC, ldap_syslog_level, "%s syncprov_add_slog: "
"updating mincsn for sid=%d csn=%s to %s\n",
op->o_log_prefix, se->se_sid, sl->sl_mincsn[i].bv_val, se->se_csn.bv_val );
ber_bvreplace( &sl->sl_mincsn[i], &se->se_csn );
"expiring csn=%s from sessionlog (sessionlog size=%d)\n",
op->o_log_prefix, se->se_csn.bv_val, sl->sl_num );
for ( i=0; i<sl->sl_numcsns; i++ )
if ( sl->sl_sids[i] >= se->se_sid )
break;
if ( i == sl->sl_numcsns || sl->sl_sids[i] != se->se_sid ) {
Debug( LDAP_DEBUG_SYNC, "%s syncprov_add_slog: "
"adding csn=%s to mincsn\n",
op->o_log_prefix, se->se_csn.bv_val, 0 );
slap_insert_csn_sids( (struct sync_cookie *)sl,
i, se->se_sid, &se->se_csn );
} else {
Log4( LDAP_DEBUG_SYNC, ldap_syslog_level, "%s syncprov_add_slog: "
"updating mincsn for sid=%d csn=%s to %s\n",
op->o_log_prefix, se->se_sid, sl->sl_mincsn[i].bv_val, se->se_csn.bv_val );
ber_bvreplace( &sl->sl_mincsn[i], &se->se_csn );
}
tavl_delete( &sl->sl_entries, se, syncprov_sessionlog_cmp );
ch_free( se );
edge = next;
sl->sl_num--;
}
ch_free( se );
sl->sl_num--;
}
}
ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wunlock( &sl->sl_mutex );
}
}
......@@ -1701,17 +1702,18 @@ playlog_cb( Operation *op, SlapReply *rs )
/* enter with sl->sl_mutex locked, release before returning */
static void
syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
sync_control *srs, BerVarray ctxcsn, int numcsns, int *sids )
sync_control *srs, BerVarray ctxcsn, int numcsns, int *sids,
struct berval *mincsn )
{
slap_overinst *on = (slap_overinst *)op->o_bd->bd_info;
slog_entry *se;
int i, j, ndel, num, nmods, mmods;
Avlnode *entry;
char cbuf[LDAP_PVT_CSNSTR_BUFSIZE];
BerVarray uuids;
struct berval delcsn[2];
if ( !sl->sl_num ) {
ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wunlock( &sl->sl_mutex );
return;
}
......@@ -1719,7 +1721,7 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
i = 0;
nmods = 0;
sl->sl_playing++;
ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wunlock( &sl->sl_mutex );
uuids = op->o_tmpalloc( (num+1) * sizeof( struct berval ) +
num * UUID_LEN, op->o_tmpmemctx );
......@@ -1729,13 +1731,31 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
delcsn[0].bv_val = cbuf;
BER_BVZERO(&delcsn[1]);
ldap_pvt_thread_rdwr_rlock( &sl->sl_mutex );
/* Make a copy of the relevant UUIDs. Put the Deletes up front
* and everything else at the end. Do this first so we can
* unlock the list mutex.
* let the write side manage the sessionlog again.
*/
for ( se=sl->sl_head; se; se=se->se_next ) {
assert( sl->sl_entries );
/* Find first relevant log entry. If greater than mincsn, backtrack one entry */
{
slog_entry te = {0};
te.se_csn = *mincsn;
entry = tavl_find3( sl->sl_entries, &te, syncprov_sessionlog_cmp, &ndel );
}
if ( ndel > 0 && entry )
entry = tavl_next( entry, TAVL_DIR_LEFT );
/* if none, just start at beginning */
if ( !entry )
entry = tavl_end( sl->sl_entries, TAVL_DIR_LEFT );
do {
slog_entry *se = entry->avl_data;
int k;
/* Make sure writes can still make progress */
ldap_pvt_thread_rdwr_runlock( &sl->sl_mutex );
ndel = 1;
for ( k=0; k<srs->sr_state.numcsns; k++ ) {
if ( se->se_sid == srs->sr_state.sids[k] ) {
......@@ -1744,6 +1764,7 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
}
}
if ( ndel <= 0 ) {
ldap_pvt_thread_rdwr_rlock( &sl->sl_mutex );
continue;
}
ndel = 0;
......@@ -1756,6 +1777,7 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
if ( ndel > 0 ) {
Debug( LDAP_DEBUG_SYNC, "%s syncprov_playlog: "
"cmp %d, too new\n", op->o_log_prefix, ndel, 0 );
ldap_pvt_thread_rdwr_rlock( &sl->sl_mutex );
break;
}
if ( se->se_tag == LDAP_REQ_DELETE ) {
......@@ -1765,8 +1787,10 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
delcsn[0].bv_len = se->se_csn.bv_len;
delcsn[0].bv_val[delcsn[0].bv_len] = '\0';
} else {
if ( se->se_tag == LDAP_REQ_ADD )
if ( se->se_tag == LDAP_REQ_ADD ) {
ldap_pvt_thread_rdwr_rlock( &sl->sl_mutex );
continue;
}
nmods++;
j = num - nmods;
}
......@@ -1775,7 +1799,7 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
uuids[j].bv_len = UUID_LEN;
if ( LogTest( LDAP_DEBUG_SYNC ) ) {
char uuidstr[40];
char uuidstr[40] = {};
lutil_uuidstr_from_normalized( uuids[j].bv_val, uuids[j].bv_len,
uuidstr, 40 );
Log4( LDAP_DEBUG_SYNC, ldap_syslog_level, "%s syncprov_playlog: "
......@@ -1783,10 +1807,12 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
op->o_log_prefix, se->se_tag == LDAP_REQ_DELETE ? "deleted" : "modified",
uuidstr, delcsn[0].bv_len ? delcsn[0].bv_val : "(null)" );
}
}
ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_rlock( &sl->sl_mutex );
} while ( (entry = tavl_next( entry, TAVL_DIR_RIGHT )) != NULL );
ldap_pvt_thread_rdwr_runlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wlock( &sl->sl_mutex );
sl->sl_playing--;
ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wunlock( &sl->sl_mutex );
ndel = i;
......@@ -2826,7 +2852,7 @@ no_change: if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
sl=si->si_logs;
if ( sl ) {
int do_play = 0;
ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wlock( &sl->sl_mutex );
/* Are there any log entries, and is the consumer state
* present in the session log?
*/
......@@ -2853,9 +2879,9 @@ no_change: if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
if ( do_play ) {
do_present = 0;
/* mutex is unlocked in playlog */
syncprov_playlog( op, rs, sl, srs, ctxcsn, numcsns, sids );
syncprov_playlog( op, rs, sl, srs, ctxcsn, numcsns, sids, &mincsn );
} else {
ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
ldap_pvt_thread_rdwr_wunlock( &sl->sl_mutex );
}
}
/* Is the CSN still present in the database? */
......@@ -3185,7 +3211,7 @@ sp_cf_gen(ConfigArgs *c)
sl = si->si_logs;
if ( !sl ) {
sl = ch_calloc( 1, sizeof( sessionlog ));
ldap_pvt_thread_mutex_init( &sl->sl_mutex );
ldap_pvt_thread_rdwr_init( &sl->sl_mutex );
si->si_logs = sl;
}
sl->sl_size = size;
......@@ -3490,19 +3516,14 @@ syncprov_db_destroy(
if ( si ) {
if ( si->si_logs ) {
sessionlog *sl = si->si_logs;
slog_entry *se = sl->sl_head;
while ( se ) {
slog_entry *se_next = se->se_next;
ch_free( se );
se = se_next;
}
tavl_free( sl->sl_entries, (AVL_FREE)ch_free );
if ( sl->sl_mincsn )
ber_bvarray_free( sl->sl_mincsn );
if ( sl->sl_sids )
ch_free( sl->sl_sids );
ldap_pvt_thread_mutex_destroy(&si->si_logs->sl_mutex);
ldap_pvt_thread_rdwr_destroy(&si->si_logs->sl_mutex);
ch_free( si->si_logs );
}
if ( si->si_ctxcsn )
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment