generic sharding : setup metadata and remote tables. Create/delete

Signed-off-by: Aakash Arayambeth <[email protected]>

Author: Aakash Arayambeth

Diff for: bbinc/cdb2_constants.h

@@ -36,6 +36,8 @@
 #define MAX_SPNAME MAXTABLELEN
 #define MAX_SPVERSION_LEN 80
 #define MAXTABLELEN 32
+#define MAXPARTITIONS 32
+#define MAXPARTITIONLEN (1 + (MAX_DBNAME_LENGTH + MAXTABLELEN)) /* partitions of the form <database>.<table> */
 #define MAXTAGLEN 64
 #define REPMAX 32
 /* Maximum buffer length for generated key name. */

Diff for: bbinc/consistent_hash.h

@@ -15,6 +15,11 @@ enum ch_err {
     CH_ERR_DUP = 4
 };
 
+enum ch_hash_func_type {
+    CH_HASH_SHA = 1,
+    CH_HASH_MD5 = 2,
+    CH_HASH_CRC = 3
+};
 struct consistent_hash_node {
     uint8_t *data;
     size_t data_len;

Diff for: bdb/bdb_schemachange.c

@@ -353,6 +353,12 @@ int bdb_llog_partition(bdb_state_type *bdb_state, tran_type *tran, char *name,
     return bdb_llog_scdone_tran(bdb_state, views, tran, name, strlen(name) + 1,
                                 bdberr);
 }
+
+int bdb_llog_hash_partition(bdb_state_type *bdb_state, tran_type *tran, char *name, int *bdberr)
+{
+    return bdb_llog_scdone_tran(bdb_state, hash_views, tran, name, strlen(name) + 1, bdberr);
+}
+
 int bdb_llog_luareload(bdb_state_type *bdb_state, int wait, int *bdberr)
 {
     return bdb_llog_scdone(bdb_state, luareload, NULL, 0, wait, bdberr);

Diff for: bdb/bdb_schemachange.h

@@ -51,7 +51,8 @@ typedef enum scdone {
     add_queue_file,          // 22
     del_queue_file,          // 23
     alias_table,             // 24
-    alias                    // 25
+    alias,                   // 25
+    hash_views               // 25
 } scdone_t;
 
 #define BDB_BUMP_DBOPEN_GEN(type, msg) \
@@ -73,6 +74,7 @@ int bdb_llog_views(bdb_state_type *bdb_state, char *name, int wait,
                    int *bdberr);
 int bdb_llog_partition(bdb_state_type *bdb_state, tran_type *tran, char *name,
                        int *bdberr);
+int bdb_llog_hash_partition(bdb_state_type *bdb_state, tran_type *tran, char *name, int *bdberr);
 int bdb_llog_rowlocks(bdb_state_type *, scdone_t, int *bdberr);
 int bdb_llog_genid_format(bdb_state_type *, scdone_t, int *bdberr);
 int bdb_reload_rowlocks(bdb_state_type *, scdone_t, int *bdberr);

Diff for: db/CMakeLists.txt

@@ -110,6 +110,7 @@ set(src
   machclass.c
   osqluprec.c
   macc_glue.c
+  hash_partition.c
   ${PROJECT_BINARY_DIR}/protobuf/bpfunc.pb-c.c
   ${PROJECT_SOURCE_DIR}/tools/cdb2_dump/cdb2_dump.c
   ${PROJECT_SOURCE_DIR}/tools/cdb2_load/cdb2_load.c

Diff for: db/comdb2.c

@@ -492,7 +492,7 @@ double gbl_sql_cost_error_threshold = -1;
 
 int gbl_parallel_recovery_threads = 0;
 
-int gbl_fdb_resolve_local = 0;
+int gbl_fdb_resolve_local = 1;
 int gbl_fdb_allow_cross_classes = 0;
 uint64_t gbl_sc_headroom = 10;
 /*---COUNTS---*/
@@ -816,6 +816,7 @@ int gbl_hostname_refresh_time = 60;
 int gbl_pstack_self = 1;
 
 char *gbl_cdb2api_policy_override = NULL;
+int gbl_create_remote_tables = 0;
 
 int close_all_dbs_tran(tran_type *tran);
 
@@ -2410,6 +2411,15 @@ int llmeta_load_timepart(struct dbenv *dbenv)
     return thedb->timepart_views ? 0 : -1;
 }
 
+int llmeta_load_hash_partitions(struct dbenv *dbenv)
+{
+    logmsg(LOGMSG_INFO, "Loading hash-based partitions\n");
+    Pthread_rwlock_init(&hash_partition_lk, NULL);
+    dbenv->hash_partition_views = hash_create_all_views();
+
+    return dbenv->hash_partition_views ? 0 : -1;
+}
+
 /* replace the table names and dbnums saved in the low level meta table with the
  * ones in the dbenv.  returns 0 on success and anything else otherwise */
 int llmeta_set_tables(tran_type *tran, struct dbenv *dbenv)
@@ -4143,6 +4153,12 @@ static int init(int argc, char **argv)
             unlock_schema_lk();
             return -1;
         }
+        
+        if (llmeta_load_hash_partitions(thedb)) {
+            logmsg(LOGMSG_ERROR, "could not load hash partitions\n");
+            unlock_schema_lk();
+            return -1;
+        }
 
         if (llmeta_load_queues(thedb)) {
             logmsg(LOGMSG_FATAL, "could not load queues from the low level meta "
@@ -6342,7 +6358,10 @@ int comdb2_reload_schemas(void *dbenv, void *inlsn)
         logmsg(LOGMSG_ERROR, "could not load time partitions\n");
         abort();
     }
-
+    if (llmeta_load_hash_partitions(thedb)) {
+        logmsg(LOGMSG_FATAL, "could not load mod based shards\n");
+        abort();
+    }
     if ((rc = bdb_get_rowlocks_state(&rlstate, tran, &bdberr)) != 0) {
         logmsg(LOGMSG_ERROR, "Get rowlocks llmeta failed, rc=%d bdberr=%d\n",
                rc, bdberr);

Diff for: db/comdb2.h

@@ -969,11 +969,11 @@ struct dbenv {
     int incoh_notcoherent;
     uint32_t incoh_file, incoh_offset;
     timepart_views_t *timepart_views;
-
-    struct time_metric *service_time;
-    struct time_metric *queue_depth;
-    struct time_metric *concurrent_queries;
-    struct time_metric *connections;
+    hash_t *hash_partition_views;
+    struct time_metric* service_time;
+    struct time_metric* queue_depth;
+    struct time_metric* concurrent_queries;
+    struct time_metric* connections;
     struct time_metric *sql_queue_time;
     struct time_metric *handle_buf_queue_time;
     struct time_metric *watchdog_time;
@@ -3734,4 +3734,5 @@ void get_disable_skipscan_all();
 
 void get_client_origin(char *out, size_t outlen, struct sqlclntstate *clnt);
 
+extern pthread_rwlock_t hash_partition_lk;
 #endif /* !INCLUDED_COMDB2_H */

Diff for: db/db_tunables.c

@@ -523,7 +523,6 @@ int gbl_old_column_names = 1;
 int gbl_enable_sq_flattening_optimization = 1;
 int gbl_mask_internal_tunables = 1;
 int gbl_allow_readonly_runtime_mod = 0;
-
 size_t gbl_cached_output_buffer_max_bytes = 8 * 1024 * 1024; /* 8 MiB */
 int gbl_sqlite_sorterpenalty = 5;
 int gbl_file_permissions = 0660;
@@ -557,6 +556,7 @@ extern int gbl_altersc_latency_inc;
 extern int gbl_sc_history_max_rows;
 extern int gbl_sc_status_max_rows;
 extern int gbl_rep_process_pstack_time;
+extern int gbl_create_remote_tables;
 
 extern void set_snapshot_impl(snap_impl_enum impl);
 extern const char *snap_impl_str(snap_impl_enum impl);

Diff for: db/db_tunables.h

@@ -2465,4 +2465,7 @@ REGISTER_TUNABLE("sc_status_max_rows", "Max number of rows returned in comdb2_sc
                  TUNABLE_INTEGER, &gbl_sc_status_max_rows, 0, NULL, NULL, NULL, NULL);
 REGISTER_TUNABLE("rep_process_pstack_time", "pstack the server if rep_process runs longer than time specified in secs (Default: 30s)",
                  TUNABLE_INTEGER, &gbl_rep_process_pstack_time, 0, NULL, NULL, NULL, NULL);
+REGISTER_TUNABLE("create_remote_tables",
+                 "while creating a sharded table, create tables (shards) on the remote databases as well (default: off)",
+                 TUNABLE_BOOLEAN, &gbl_create_remote_tables, 0, NULL, NULL, NULL, NULL);
 #endif /* _DB_TUNABLES_H */