Simple multiversion test crashes (db-5.3.15)

Question

Hi!
I wrote a small test for a multiversion database (using the C++ API). The test starts a transaction with DB_TXN_SNAPSHOT and reads/writes a key-value pair foo=bar. It then sleeps 10 seconds to allow starting a second instance of the program. The first instance succeeds, with this output:
$ mvcctest
get key foo
put key foo
successfully wrote key foo=bar
waiting 10 seconds...
$
The second instance which I started in gdb to get the stacktrace produces this output/stacktrace:
(gdb) r
Starting program: /home/stefan/src/sandbox/bdb-trans/mvcctest
[Thread debugging using libthread_db enabled]
get key foo
put key foo
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff7b5bd01 in __memp_fput (dbmfp=0x605230, ip=0x0, pgaddr=0x0, priority=DB_PRIORITY_UNCHANGED)
at ../src/mp/mp_fput.c:125
125 infop = &dbmp->reginfo[bhp->region];
(gdb) bt
#0 0x00007ffff7b5bd01 in __memp_fput (dbmfp=0x605230, ip=0x0, pgaddr=0x0, priority=DB_PRIORITY_UNCHANGED)
at ../src/mp/mp_fput.c:125
#1 0x00007ffff79f3cce in __bam_get_root (dbc=0x605a10, root_pgno=1, slevel=1, flags=12802, stack=0x7fffffffd75c)
at ../src/btree/bt_search.c:235
#2 0x00007ffff79f3ed6 in __bam_search (dbc=0x605a10, root_pgno=1, key=0x7fffffffdbf0, flags=12802, slevel=1, recnop=0x0,
exactp=0x7fffffffd934) at ../src/btree/bt_search.c:309
#3 0x00007ffff79dce1c in __bamc_search (dbc=0x605a10, root_pgno=1, key=0x7fffffffdbf0, flags=14, exactp=0x7fffffffd934)
at ../src/btree/bt_cursor.c:2804
#4 0x00007ffff79dad0b in __bamc_put (dbc=0x605a10, key=0x7fffffffdbf0, data=0x7fffffffdc20, flags=14, pgnop=0x7fffffffd9b0)
at ../src/btree/bt_cursor.c:2143
#5 0x00007ffff7adf03f in __dbc_iput (dbc=0x605a10, key=0x7fffffffdbf0, data=0x7fffffffdc20, flags=14)
at ../src/db/db_cam.c:2136
#6 0x00007ffff7adee8d in __dbc_put (dbc=0x605a10, key=0x7fffffffdbf0, data=0x7fffffffdc20, flags=14)
at ../src/db/db_cam.c:2049
#7 0x00007ffff7ad78dd in __db_put (dbp=0x604af0, ip=0x0, txn=0x605390, key=0x7fffffffdbf0, data=0x7fffffffdc20, flags=0)
at ../src/db/db_am.c:583
#8 0x00007ffff7af20c8 in __db_put_pp (dbp=0x604af0, txn=0x605390, key=0x7fffffffdbf0, data=0x7fffffffdc20, flags=0)
at ../src/db/db_iface.c:1661
#9 0x00007ffff79bc576 in Db::put (this=0x604a50, txnid=0x605fa0, key=0x7fffffffdbf0, value=0x7fffffffdc20, flags=0)
at ../lang/cxx/cxx_db.cpp:367
#10 0x0000000000400caf in main () at mvcctest.cc:60
I am unsure if this is my fault (because the testcase does something it should not do), or if this is a Berkeley DB bug.
#include <sys/stat.h>
#include <sys/types.h>
#include <assert.h>
#include <db_cxx.h>
#include <string.h>
#include <stdlib.h>
#include <string>
using std::string;
int
main()
mkdir ("bdb", 0770);
DbEnv *db_env = new DbEnv (DB_CXX_NO_EXCEPTIONS);
int ret = db_env->open ("bdb",
DB_CREATE | /* on-demand create */
DB_INIT_MPOOL | /* shared memory buffer subsystem */
DB_INIT_TXN | /* transactions */
DB_INIT_LOG | /* logging */
DB_INIT_LOCK | /* locking */
DB_MULTIVERSION,
0);
assert (ret == 0);
db_env->set_flags (DB_MULTIVERSION, 1);
Db *db = new Db (db_env, 0);
db->set_flags (DB_DUP); // allow duplicate keys
// Open the database
u_int32_t oFlags = DB_CREATE | DB_AUTO_COMMIT | DB_MULTIVERSION; // Open flags;
db->open (NULL, // Transaction pointer
"db", // Database name
NULL, // Optional logical database name
DB_BTREE, // Database access method
oFlags, // Open flags
0); // File mode (using defaults)
DbTxn* transaction;
ret = db_env->txn_begin (NULL, &transaction, DB_TXN_SNAPSHOT);
assert (ret == 0);
char foo[4] = "foo";
char bar[4] = "bar";
Dbt key (foo, 3);
Dbt new_data (bar, 3);
Dbt data;
printf ("get key foo\n");
ret = db->get (transaction, &key, &data, 0);
if (ret == 0)
printf ("successfully read key foo\n");
printf ("put key foo\n");
ret = db->put (transaction, &key, &new_data, 0);
if (ret == 0)
printf ("successfully wrote key foo=bar\n");
printf ("waiting 10 seconds...\n");
sleep (10);
ret = transaction->commit (0);
assert (ret == 0);
ret = db->close (0);
delete db;
db = NULL;
assert (ret == 0);
ret = db_env->close (0);
delete db_env;
db_env = NULL;
assert (ret == 0);
}

Answer 1

Yes, we have the same error:
com.sleepycat.dbxml.XmlException: std::exception thrown from C++ API: bad allocation: Unknown error, errcode = INTERNAL_ERROR
     at com.sleepycat.dbxml.dbxml_javaJNI.XmlQueryExpression_execute__SWIG_1(Native Method)
     at com.sleepycat.dbxml.XmlQueryExpression.execute(XmlQueryExpression.java:93)
com.sleepycat.dbxml.XmlException, errcode = NO_MEMORY_ERROR
     at com.sleepycat.dbxml.dbxml_javaJNI.XmlQueryExpression_execute__SWIG_1(Native Method)
     at com.sleepycat.dbxml.XmlQueryExpression.execute(XmlQueryExpression.java:93)
The version we use is 2.2.13, can anyone give suggestions to resolve it?
Regards,
Jane