[t:2499] Add recover-loader-test

git-svn-id: file:///svn/toku/tokudb@19900 c7de825b-a66e-492c-adef-691d508d4ae1

src/tests/Makefile

@@ -147,6 +147,7 @@ BDB_DONTRUN_TESTS = \
 	recover-put-multiple-fdelete-all \
 	recover-put-multiple-fdelete-some \
 	recover-split-checkpoint \
+	recover-loader-test \
 	progress \
 	redirect \
         isolation-read-committed \
@@ -456,7 +457,7 @@ libs:
 
 
 
-loader-cleanup-test.tdb$(BINSUF) diskfull.tdb$(BINSUF): CPPFLAGS+=-DDONT_DEPRECATE_WRITES
+loader-cleanup-test.tdb$(BINSUF) recover-loader-test.tdb$(BINSUF) diskfull.tdb$(BINSUF): CPPFLAGS+=-DDONT_DEPRECATE_WRITES
 test_db_curs4.tdb$(BINSUF): trace.h
 test_db_curs4.bdb$(BINSUF): trace.h
 # a bunch of little tests designed to run in parallel

src/tests/recover-loader-test.c

+/* -*- mode: C; c-basic-offset: 4 -*- */
+#ident "Copyright (c) 2010 Tokutek Inc.  All rights reserved."
+#ident "$Id: loader-stress-test.c 19683 2010-04-30 18:19:17Z yfogel $"
+
+
+
+/* Purpose is to verify that when a loader crashes:
+ *  - there are no temp files remaining
+ *  - the loader-generated iname file is not present
+ *
+ * In the event of a crash, the verification of no temp files and 
+ * no loader-generated iname file is done after recovery.
+ *
+ * Mechanism:
+ * This test is derived from loader-cleanup-test, which was derived from loader-stress-test.
+ *
+ * The outline of the test is as follows:
+ *  - use loader to create table
+ *  - verify presence of temp files
+ *  - crash
+ *  - recover
+ *  - verify absence of temp files
+ *  - verify absence of unwanted iname files (old inames if committed, new inames if aborted)
+ *
+ *  
+ */
+
+
+#include "test.h"
+#include "toku_pthread.h"
+#include <db.h>
+#include <sys/stat.h>
+
+#include <sys/types.h>
+#include <dirent.h>
+
+#include "ydb-internal.h"
+
+static const int envflags = DB_INIT_LOCK | DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_TXN | DB_CREATE | DB_PRIVATE;
+
+BOOL do_test=FALSE, do_recover=FALSE;
+
+DB_ENV *env;
+enum {MAX_NAME=128};
+enum {MAX_DBS=256};
+int NUM_DBS=5;
+int NUM_ROWS=100000;
+int CHECK_RESULTS=0;
+int USE_PUTS=0;
+int INDUCE_ENOSPC=0;
+enum {MAGIC=311};
+
+
+DBT old_inames[MAX_DBS];
+DBT new_inames[MAX_DBS];
+
+
+int count_temp(char * dirname);
+void get_inames(DBT* inames, DB** dbs);
+int verify_file(char * dirname, char * filename);
+void assert_inames_missing(DBT* inames);
+int print_dir(char * dirname);
+
+// return number of temp files
+int
+count_temp(char * dirname) {
+    int n = 0;
+    
+    DIR * dir = opendir(dirname);
+    
+    struct dirent *ent;
+    while ((ent=readdir(dir))) {
+	if (ent->d_type==DT_REG && strncmp(ent->d_name, "temp", 4)==0) {
+	    n++;
+	}
+    }
+    closedir(dir);
+    return n;
+}
+
+// print contents of directory
+int
+print_dir(char * dirname) {
+    int n;
+
+    DIR * dir = opendir(dirname);
+    
+    struct dirent *ent;
+    while ((ent=readdir(dir))) {
+	if (ent->d_type==DT_REG) {
+	    n++;
+	    printf("File: %s\n", ent->d_name);
+	}
+    }
+    closedir(dir);
+    return n;
+}
+
+
+
+// return non-zero if file exists
+int 
+verify_file(char * dirname, char * filename) {
+    int n = 0;
+    DIR * dir = opendir(dirname);
+    
+    struct dirent *ent;
+    while ((ent=readdir(dir))) {
+	if (ent->d_type==DT_REG && strcmp(ent->d_name, filename)==0) {
+	    n++;
+	}
+    }
+    closedir(dir);
+    return n;
+}
+
+void
+get_inames(DBT* inames, DB** dbs) {
+    int i;
+    for (i = 0; i < NUM_DBS; i++) {
+	DBT dname;
+	char * dname_str = dbs[i]->i->dname;
+	dbt_init(&dname, dname_str, sizeof(dname_str));
+	dbt_init(&(inames[i]), NULL, 0);
+	inames[i].flags |= DB_DBT_MALLOC;
+	int r = env->get_iname(env, &dname, &inames[i]);
+	CKERR(r);
+	char * iname_str = (char*) (inames[i].data);
+	if (verbose) printf("dname = %s, iname = %s\n", dname_str, iname_str);
+    }
+}
+
+
+void 
+assert_inames_missing(DBT* inames) {
+    int i;
+    char * dir = env->i->real_data_dir;
+    for (i=0; i<NUM_DBS; i++) {
+	char * iname = inames[i].data;
+	int r = verify_file(dir, iname);
+	if (r) {
+	    printf("File %s exists, but it should not\n", iname);
+	}
+	assert(r == 0);
+	if (verbose) printf("File has been properly deleted: %s\n", iname);
+    }
+}
+
+
+#if 0
+void print_inames(DB** dbs);
+void
+print_inames(DB** dbs) {
+    int i;
+    for (i = 0; i < NUM_DBS; i++) {
+	DBT dname;
+	DBT iname;
+	char * dname_str = dbs[i]->i->dname;
+	dbt_init(&dname, dname_str, sizeof(dname_str));
+	dbt_init(&iname, NULL, 0);
+	iname.flags |= DB_DBT_MALLOC;
+	int r = env->get_iname(env, &dname, &iname);
+	CKERR(r);
+	char * iname_str = (char*)iname.data;
+	if (verbose) printf("dname = %s, iname = %s\n", dname_str, iname_str);
+	int n = verify_file(env->i->real_data_dir, iname_str);
+	assert(n == 1);
+	toku_free(iname.data);
+    }
+}
+#endif
+
+
+//
+//   Functions to create unique key/value pairs, row generators, checkers, ... for each of NUM_DBS
+//
+
+//   a is the bit-wise permute table.  For DB[i], permute bits as described in a[i] using 'twiddle32'
+// inv is the inverse bit-wise permute of a[].  To get the original value from a twiddled value, twiddle32 (again) with inv[]
+int   a[MAX_DBS][32];
+int inv[MAX_DBS][32];
+
+#if defined(__cilkplusplus) || defined (__cplusplus)
+extern "C" {
+#endif
+
+// rotate right and left functions
+static inline unsigned int rotr32(const unsigned int x, const unsigned int num) {
+    const unsigned int n = num % 32;
+    return (x >> n) | ( x << (32 - n));
+}
+static inline unsigned int rotl32(const unsigned int x, const unsigned int num) {
+    const unsigned int n = num % 32;
+    return (x << n) | ( x >> (32 - n));
+}
+
+static void generate_permute_tables(void) {
+    int i, j, tmp;
+    for(int db=0;db<MAX_DBS;db++) {
+        for(i=0;i<32;i++) {
+            a[db][i] = i;
+        }
+        for(i=0;i<32;i++) {
+            j = random() % (i + 1);
+            tmp = a[db][j];
+            a[db][j] = a[db][i];
+            a[db][i] = tmp;
+        }
+//        if(db < NUM_DBS){ printf("a[%d] = ", db); for(i=0;i<32;i++) { printf("%2d ", a[db][i]); } printf("\n");}
+        for(i=0;i<32;i++) {
+            inv[db][a[db][i]] = i;
+        }
+    }
+}
+
+// permute bits of x based on permute table bitmap
+static unsigned int twiddle32(unsigned int x, int db)
+{
+    unsigned int b = 0;
+    for(int i=0;i<32;i++) {
+        b |= (( x >> i ) & 1) << a[db][i];
+    }
+    return b;
+}
+
+// generate val from key, index
+static unsigned int generate_val(int key, int i) {
+    return rotl32((key + MAGIC), i);
+}
+
+// There is no handlerton in this test, so this function is a local replacement
+// for the handlerton's generate_row_for_put().
+static int put_multiple_generate(DB *dest_db, DB *src_db, DBT *dest_key, DBT *dest_val, const DBT *src_key, const DBT *src_val, void *extra) {
+
+    src_db = src_db;
+    extra = extra;
+
+    uint32_t which = *(uint32_t*)dest_db->app_private;
+
+    if ( which == 0 ) {
+        if (dest_key->flags==DB_DBT_REALLOC) {
+            if (dest_key->data) toku_free(dest_key->data);
+            dest_key->flags = 0;
+            dest_key->ulen  = 0;
+        }
+        if (dest_val->flags==DB_DBT_REALLOC) {
+            if (dest_val->data) toku_free(dest_val->data);
+            dest_val->flags = 0;
+            dest_val->ulen  = 0;
+        }
+        dbt_init(dest_key, src_key->data, src_key->size);
+        dbt_init(dest_val, src_val->data, src_val->size);
+    }
+    else {
+        assert(dest_key->flags==DB_DBT_REALLOC);
+        if (dest_key->ulen < sizeof(unsigned int)) {
+            dest_key->data = toku_xrealloc(dest_key->data, sizeof(unsigned int));
+            dest_key->ulen = sizeof(unsigned int);
+        }
+        assert(dest_val->flags==DB_DBT_REALLOC);
+        if (dest_val->ulen < sizeof(unsigned int)) {
+            dest_val->data = toku_xrealloc(dest_val->data, sizeof(unsigned int));
+            dest_val->ulen = sizeof(unsigned int);
+        }
+        unsigned int *new_key = (unsigned int *)dest_key->data;
+        unsigned int *new_val = (unsigned int *)dest_val->data;
+
+        *new_key = twiddle32(*(unsigned int*)src_key->data, which);
+        *new_val = generate_val(*(unsigned int*)src_key->data, which);
+
+        dest_key->size = sizeof(unsigned int);
+        dest_val->size = sizeof(unsigned int);
+        //data is already set above
+    }
+
+//    printf("dest_key.data = %d\n", *(int*)dest_key->data);
+//    printf("dest_val.data = %d\n", *(int*)dest_val->data);
+
+    return 0;
+}
+
+#if defined(__cilkplusplus) || defined(__cplusplus)
+} // extern "C"
+#endif
+
+static void *expect_poll_void = &expect_poll_void;
+
+static int poll_function (void *UU(extra), float UU(progress)) {
+    toku_hard_crash_on_purpose();
+    return -1;
+}
+
+static void test_loader(DB **dbs)
+{
+    int r;
+    DB_TXN    *txn;
+    DB_LOADER *loader;
+    uint32_t db_flags[MAX_DBS];
+    uint32_t dbt_flags[MAX_DBS];
+    for(int i=0;i<MAX_DBS;i++) { 
+        db_flags[i] = DB_NOOVERWRITE; 
+        dbt_flags[i] = 0;
+    }
+    uint32_t loader_flags = USE_PUTS; // set with -p option
+
+    if (verbose) printf("old inames:\n");
+    get_inames(old_inames, dbs);
+    
+    // create and initialize loader
+    r = env->txn_begin(env, NULL, &txn, 0);                                                               
+    CKERR(r);
+    r = env->create_loader(env, txn, &loader, dbs[0], NUM_DBS, dbs, db_flags, dbt_flags, loader_flags);
+    CKERR(r);
+    r = loader->set_error_callback(loader, NULL, NULL);
+    CKERR(r);
+    r = loader->set_poll_function(loader, poll_function, expect_poll_void);
+    CKERR(r);
+
+    printf("USE_PUTS = %d\n", USE_PUTS);
+    if (verbose) printf("new inames:\n");
+    get_inames(new_inames, dbs);
+
+    // using loader->put, put values into DB
+    DBT key, val;
+    unsigned int k, v;
+    for(int i=1;i<=NUM_ROWS;i++) {
+        k = i;
+        v = generate_val(i, 0);
+        dbt_init(&key, &k, sizeof(unsigned int));
+        dbt_init(&val, &v, sizeof(unsigned int));
+        r = loader->put(loader, &key, &val);
+	CKERR(r);
+        if ( CHECK_RESULTS || verbose) { if((i%10000) == 0){printf("."); fflush(stdout);} }
+    }
+    if( CHECK_RESULTS || verbose ) {printf("\n"); fflush(stdout);}        
+        
+    printf("Data dir is %s\n", env->i->real_data_dir);
+    int n = count_temp(env->i->real_data_dir);
+    printf("Num temp files = %d\n", n);
+    assert(n);  // test is useless unless at least one temp file is created
+
+    printf("Contents of data dir:\n");
+    print_dir(env->i->real_data_dir);
+    printf("closing, will crash\n"); fflush(stdout);
+    r = loader->close(loader);
+    printf("Should never return from loader->close()\n");  fflush(stdout);
+    assert(0);
+
+}
+
+
+static void run_test(void)
+{
+    int r;
+
+    r = system("rm -rf " ENVDIR);                                                                             CKERR(r);
+    r = toku_os_mkdir(ENVDIR, S_IRWXU+S_IRWXG+S_IRWXO);                                                       CKERR(r);
+
+    r = db_env_create(&env, 0);                                                                               CKERR(r);
+    r = env->set_default_bt_compare(env, uint_dbt_cmp);                                                       CKERR(r);
+    r = env->set_default_dup_compare(env, uint_dbt_cmp);                                                      CKERR(r);
+    r = env->set_generate_row_callback_for_put(env, put_multiple_generate);
+    CKERR(r);
+//    int envflags = DB_INIT_LOCK | DB_INIT_MPOOL | DB_INIT_TXN | DB_CREATE | DB_PRIVATE;
+    r = env->open(env, ENVDIR, envflags, S_IRWXU+S_IRWXG+S_IRWXO);                                            CKERR(r);
+    env->set_errfile(env, stderr);
+    //Disable auto-checkpointing
+    r = env->checkpointing_set_period(env, 0);                                                                CKERR(r);
+
+    DBT desc;
+    dbt_init(&desc, "foo", sizeof("foo"));
+    char name[MAX_NAME*2];
+
+    DB **dbs = (DB**)toku_malloc(sizeof(DB*) * NUM_DBS);
+    assert(dbs != NULL);
+    int idx[MAX_DBS];
+    for(int i=0;i<NUM_DBS;i++) {
+        idx[i] = i;
+        r = db_create(&dbs[i], env, 0);                                                                       CKERR(r);
+        r = dbs[i]->set_descriptor(dbs[i], 1, &desc, abort_on_upgrade);                                       CKERR(r);
+        dbs[i]->app_private = &idx[i];
+        snprintf(name, sizeof(name), "db_%04x", i);
+        r = dbs[i]->open(dbs[i], NULL, name, NULL, DB_BTREE, DB_CREATE, 0666);                                CKERR(r);
+    }
+
+    generate_permute_tables();
+
+    test_loader(dbs);
+    printf("Should never return from test_loader\n");  fflush(stdout);
+    assert(0);
+}
+
+// ------------ infrastructure ----------
+static void do_args(int argc, char * const argv[]);
+
+
+
+static void run_recover (void) {
+
+    // Recovery starts from oldest_living_txn, which is older than any inserts done in run_test,
+    // so recovery always runs over the entire log.
+
+    // run recovery
+    int r = db_env_create(&env, 0);                                                         CKERR(r);
+    r = env->open(env, ENVDIR, envflags + DB_RECOVER, S_IRWXU+S_IRWXG+S_IRWXO);         CKERR(r); 
+
+    // now verify contents of data_dir, should be no temp files, no loader-created iname files
+    print_dir(env->i->real_data_dir);
+
+    int n = count_temp(env->i->real_data_dir);
+    printf("Num temp files = %d\n", n);
+    assert(n==0);  // There should be no temp files remaining after recovery
+
+    r = env->close(env, 0);                                                             CKERR(r);
+    exit(0);
+
+}
+
+int test_main(int argc, char * const *argv) {
+    do_args(argc, argv);
+
+    if (do_test) {
+	printf("\n\n perform test, crash\n");
+	fflush(stdout);
+	run_test();
+    }
+    else if (do_recover) {
+	printf("\n\n perform recovery\n");
+	run_recover();
+    }
+    else {
+	printf("\n\n BOGUS!\n");
+	assert(0);
+    }
+
+    return 0;
+}
+
+static void do_args(int argc, char * const argv[]) {
+    int resultcode;
+    char *cmd = argv[0];
+    argc--; argv++;
+    while (argc>0) {
+	if (strcmp(argv[0], "-v")==0) {
+	    verbose++;
+	} else if (strcmp(argv[0],"-q")==0) {
+	    verbose--;
+	    if (verbose<0) verbose=0;
+        } else if (strcmp(argv[0], "-h")==0) {
+	    resultcode=0;
+	do_usage:
+	    fprintf(stderr, "Usage: -h -c -d <num_dbs> -r <num_rows>\n%s\n", cmd);
+	    exit(resultcode);
+        } else if (strcmp(argv[0], "-d")==0) {
+            argc--; argv++;
+            NUM_DBS = atoi(argv[0]);
+            if ( NUM_DBS > MAX_DBS ) {
+                fprintf(stderr, "max value for -d field is %d\n", MAX_DBS);
+                resultcode=1;
+                goto do_usage;
+            }
+        } else if (strcmp(argv[0], "-r")==0) {
+            argc--; argv++;
+            NUM_ROWS = atoi(argv[0]);
+        } else if (strcmp(argv[0], "-c")==0) {
+            CHECK_RESULTS = 1;
+        } else if (strcmp(argv[0], "-p")==0) {
+            USE_PUTS = LOADER_USE_PUTS;
+	    printf("Using puts\n");
+        } else if (strcmp(argv[0], "-e")==0) {
+	    INDUCE_ENOSPC = 1;
+	    printf("Using enospc\n");
+
+	} else if (strcmp(argv[0], "--test")==0) {
+	    do_test=TRUE;
+        } else if (strcmp(argv[0], "--recover") == 0) {
+            do_recover=TRUE;
+
+	} else {
+	    fprintf(stderr, "Unknown arg: %s\n", argv[0]);
+	    resultcode=1;
+	    goto do_usage;
+	}
+	argc--;
+	argv++;
+    }
+}