From 460016428cf27484c333227f534c2e2f73a37fb1 Mon Sep 17 00:00:00 2001 From: Milind Changire Date: Sat, 15 Oct 2016 10:49:19 +0530 Subject: cluster/tier: handle fast demotions Demote files on priority if hi-watermark has been breached and continue to demote until the watermark drops below hi-watermark. Monitor watermark more frequently. Trigger demotion as soon as hi-watermark is breached. Add cluster.tier-emergency-demote-query-limit option to limit number of files returned from the database query for every iteration of tier_migrate_using_query_file(). If watermark hasn't dropped below hi-watermark during the first iteration, the next iteration will be triggered approximately 1 second after tier_demote() returns to the main tiering loop. Update changetimerecorder xlator to handle query for emergency demote mode. Add tier-ctr-interface.h: Move tier and ctr interface specific macros and struct definition from libglusterfs/src/gfdb/gfdb_data_store.h to new header libglusterfs/src/tier-ctr-interface.h Change-Id: If56af78c6c81d37529b9b6e65ae606ba5c99a811 BUG: 1366648 Signed-off-by: Milind Changire Reviewed-on: http://review.gluster.org/15158 Smoke: Gluster Build System CentOS-regression: Gluster Build System NetBSD-regression: NetBSD Build System Reviewed-by: Dan Lambright --- libglusterfs/src/Makefile.am | 3 +- libglusterfs/src/gfdb/gfdb_data_store.c | 8 +++-- libglusterfs/src/gfdb/gfdb_data_store.h | 48 ++++++--------------------- libglusterfs/src/gfdb/gfdb_data_store_types.h | 8 ++++- libglusterfs/src/gfdb/gfdb_sqlite3.c | 37 +++++++++++++++++---- libglusterfs/src/gfdb/gfdb_sqlite3.h | 3 +- libglusterfs/src/tier-ctr-interface.h | 44 ++++++++++++++++++++++++ 7 files changed, 102 insertions(+), 49 deletions(-) create mode 100644 libglusterfs/src/tier-ctr-interface.h (limited to 'libglusterfs') diff --git a/libglusterfs/src/Makefile.am b/libglusterfs/src/Makefile.am index 20163da769c..e9e690ee4bd 100644 --- a/libglusterfs/src/Makefile.am +++ b/libglusterfs/src/Makefile.am @@ -61,7 +61,8 @@ noinst_HEADERS = unittest/unittest.h \ $(CONTRIBDIR)/rbtree/rb.h \ $(CONTRIBDIR)/mount/mntent_compat.h \ $(CONTRIBDIR)/libexecinfo/execinfo_compat.h \ - $(CONTRIBDIR)/timer-wheel/timer-wheel.h + $(CONTRIBDIR)/timer-wheel/timer-wheel.h \ + tier-ctr-interface.h if !HAVE_LIBUUID # FIXME: unbundle libuuid, see compat-uuid.h. diff --git a/libglusterfs/src/gfdb/gfdb_data_store.c b/libglusterfs/src/gfdb/gfdb_data_store.c index cb567503fa3..7074c4a51c2 100644 --- a/libglusterfs/src/gfdb/gfdb_data_store.c +++ b/libglusterfs/src/gfdb/gfdb_data_store.c @@ -481,12 +481,14 @@ compact_db (gfdb_conn_node_t *_conn_node, gf_boolean_t _compact_active, * for every record found * _query_cbk_args : Custom argument passed for the call back * function query_callback + * query_limit : number to limit number of rows returned by the query * Returns : if successful return 0 or * -ve value in case of failure*/ int find_all (gfdb_conn_node_t *_conn_node, gf_query_callback_t query_callback, - void *_query_cbk_args) + void *_query_cbk_args, + int query_limit) { int ret = 0; gfdb_db_operations_t *db_operations_t = NULL; @@ -500,7 +502,8 @@ find_all (gfdb_conn_node_t *_conn_node, if (db_operations_t->find_all_op) { ret = db_operations_t->find_all_op (gf_db_connection, query_callback, - _query_cbk_args); + _query_cbk_args, + query_limit); if (ret) { gf_msg (GFDB_DATA_STORE, GF_LOG_ERROR, 0, LG_MSG_FIND_OP_FAILED, "Find all operation " @@ -851,6 +854,7 @@ void get_gfdb_methods (gfdb_methods_t *methods) { methods->init_db = init_db; methods->fini_db = fini_db; + methods->find_all = find_all; methods->find_unchanged_for_time = find_unchanged_for_time; methods->find_recently_changed_files = find_recently_changed_files; methods->find_unchanged_for_time_freq = find_unchanged_for_time_freq; diff --git a/libglusterfs/src/gfdb/gfdb_data_store.h b/libglusterfs/src/gfdb/gfdb_data_store.h index 0aac4611153..beb954c190a 100644 --- a/libglusterfs/src/gfdb/gfdb_data_store.h +++ b/libglusterfs/src/gfdb/gfdb_data_store.h @@ -20,42 +20,6 @@ #include "gfdb_data_store_types.h" -#define GFDB_IPC_CTR_KEY "gfdb.ipc-ctr-op" - -/* - * CTR IPC OPERATIONS - * - * - */ -#define GFDB_IPC_CTR_QUERY_OPS "gfdb.ipc-ctr-query-op" -#define GFDB_IPC_CTR_CLEAR_OPS "gfdb.ipc-ctr-clear-op" -#define GFDB_IPC_CTR_GET_DB_PARAM_OPS "gfdb.ipc-ctr-get-db-parm" -#define GFDB_IPC_CTR_GET_DB_VERSION_OPS "gfdb.ipc-ctr-get-db-version" -#define GFDB_IPC_CTR_SET_COMPACT_PRAGMA "gfdb.ipc-ctr-set-compact-pragma" -/* - * CTR IPC INPUT/OUTPUT - * - * - */ -#define GFDB_IPC_CTR_GET_QFILE_PATH "gfdb.ipc-ctr-get-qfile-path" -#define GFDB_IPC_CTR_GET_QUERY_PARAMS "gfdb.ipc-ctr-get-query-parms" -#define GFDB_IPC_CTR_RET_QUERY_COUNT "gfdb.ipc-ctr-ret-rec-count" -#define GFDB_IPC_CTR_GET_DB_KEY "gfdb.ipc-ctr-get-params-key" -#define GFDB_IPC_CTR_RET_DB_VERSION "gfdb.ipc-ctr-ret-db-version" - -/* - * gfdb ipc ctr params for query - * - * - */ -typedef struct gfdb_ipc_ctr_params { - gf_boolean_t is_promote; - int write_freq_threshold; - int read_freq_threshold; - gfdb_time_t time_stamp; -} gfdb_ipc_ctr_params_t; - - /* GFDB Connection Node: * ~~~~~~~~~~~~~~~~~~~~ * Represents the connection to the database while using libgfdb @@ -146,11 +110,20 @@ delete_record(gfdb_conn_node_t *, gfdb_db_record_t *gfdb_db_record); * for every record found * _query_cbk_args : Custom argument passed for the call back * function query_callback + * query_limit : 0 - umlimited, + * any positive value - adds the LIMIT clause + * to the SQL query + * * Returns : if successful return 0 or * -ve value in case of failure*/ int find_all(gfdb_conn_node_t *, gf_query_callback_t query_callback, - void *_query_cbk_args); + void *_query_cbk_args, + int query_limit); +typedef int (*find_all_t) (gfdb_conn_node_t *, + gf_query_callback_t query_callback, + void *_query_cbk_args, + int query_limit); @@ -368,6 +341,7 @@ typedef int (*compact_db_t)(gfdb_conn_node_t *db_conn, typedef struct gfdb_methods_s { init_db_t init_db; fini_db_t fini_db; + find_all_t find_all; find_unchanged_for_time_t find_unchanged_for_time; find_recently_changed_files_t find_recently_changed_files; find_unchanged_for_time_freq_t find_unchanged_for_time_freq; diff --git a/libglusterfs/src/gfdb/gfdb_data_store_types.h b/libglusterfs/src/gfdb/gfdb_data_store_types.h index d0c96370eb8..02b7aa0fd33 100644 --- a/libglusterfs/src/gfdb/gfdb_data_store_types.h +++ b/libglusterfs/src/gfdb/gfdb_data_store_types.h @@ -366,12 +366,18 @@ typedef int * for every record found * _query_cbk_args : Custom argument passed for the call back * function query_callback + * query_limit : 0 - list all files + * positive value - add the LIMIT clause to + * the SQL query to limit the number of records + * returned + * * Returns : if successful return 0 or * -ve value in case of failure*/ typedef int (*gfdb_find_all_t)(void *db_conn, gf_query_callback_t query_callback, - void *_cbk_args); + void *_cbk_args, + int query_limit); diff --git a/libglusterfs/src/gfdb/gfdb_sqlite3.c b/libglusterfs/src/gfdb/gfdb_sqlite3.c index 094028361c5..4284ed9a69d 100644 --- a/libglusterfs/src/gfdb/gfdb_sqlite3.c +++ b/libglusterfs/src/gfdb/gfdb_sqlite3.c @@ -632,12 +632,15 @@ gf_get_basic_query_stmt (char **out_stmt) * */ int gf_sqlite3_find_all (void *db_conn, gf_query_callback_t query_callback, - void *query_cbk_args) + void *query_cbk_args, + int query_limit) { int ret = -1; char *query_str = NULL; gf_sql_connection_t *sql_conn = db_conn; sqlite3_stmt *prep_stmt = NULL; + char *limit_query = NULL; + char *query = NULL; CHECK_SQL_CONN (sql_conn, out); GF_VALIDATE_OR_GOTO(GFDB_STR_SQLITE3, query_callback, out); @@ -647,12 +650,28 @@ gf_sqlite3_find_all (void *db_conn, gf_query_callback_t query_callback, goto out; } - ret = sqlite3_prepare (sql_conn->sqlite3_db_conn, query_str, -1, + query = query_str; + + if (query_limit > 0) { + ret = gf_asprintf (&limit_query, "%s LIMIT %d", + query, query_limit); + if (ret < 0) { + gf_msg (GFDB_STR_SQLITE3, GF_LOG_ERROR, 0, + LG_MSG_QUERY_FAILED, + "Failed creating limit query statement"); + limit_query = NULL; + goto out; + } + + query = limit_query; + } + + ret = sqlite3_prepare (sql_conn->sqlite3_db_conn, query, -1, &prep_stmt, 0); if (ret != SQLITE_OK) { gf_msg (GFDB_STR_SQLITE3, GF_LOG_ERROR, 0, - LG_MSG_PREPARE_FAILED, "Failed to prepare statement %s :" - "%s", query_str, + LG_MSG_PREPARE_FAILED, + "Failed to prepare statement %s: %s", query, sqlite3_errmsg (sql_conn->sqlite3_db_conn)); ret = -1; goto out; @@ -661,7 +680,7 @@ gf_sqlite3_find_all (void *db_conn, gf_query_callback_t query_callback, ret = gf_sql_query_function (prep_stmt, query_callback, query_cbk_args); if (ret) { gf_msg (GFDB_STR_SQLITE3, GF_LOG_ERROR, 0, LG_MSG_QUERY_FAILED, - "Failed Query %s", query_str); + "Failed Query %s", query); goto out; } @@ -669,6 +688,10 @@ gf_sqlite3_find_all (void *db_conn, gf_query_callback_t query_callback, out: sqlite3_finalize (prep_stmt); GF_FREE (query_str); + + if (limit_query) + GF_FREE (limit_query); + return ret; } @@ -1070,10 +1093,10 @@ gf_sqlite3_find_unchanged_for_time_freq (void *db_conn, GF_COL_TB_WMSEC ") >= ? ) ) )" " AND " /*Second condition: For Reads - * Files that have reaASCd wind time smaller than for_time + * Files that have read wind time smaller than for_time * OR * File that have read wind time greater than for_time, - * but write_frequency less than freq_write_cnt*/ + * but read_frequency less than freq_read_cnt*/ "( ((" GF_COL_TB_RWSEC " * " TOSTRING(GFDB_MICROSEC) " + " GF_COL_TB_RWMSEC ") < ? )" " OR " diff --git a/libglusterfs/src/gfdb/gfdb_sqlite3.h b/libglusterfs/src/gfdb/gfdb_sqlite3.h index 52b84da251c..5b55b0ace5e 100644 --- a/libglusterfs/src/gfdb/gfdb_sqlite3.h +++ b/libglusterfs/src/gfdb/gfdb_sqlite3.h @@ -258,7 +258,8 @@ int gf_sqlite3_delete (void *db_conn, gfdb_db_record_t *); /*querying modules*/ int gf_sqlite3_find_all (void *db_conn, gf_query_callback_t, - void *_query_cbk_args); + void *_query_cbk_args, + int query_limit); int gf_sqlite3_find_unchanged_for_time (void *db_conn, gf_query_callback_t query_callback, void *_query_cbk_args, diff --git a/libglusterfs/src/tier-ctr-interface.h b/libglusterfs/src/tier-ctr-interface.h new file mode 100644 index 00000000000..cfd3f8a5e5d --- /dev/null +++ b/libglusterfs/src/tier-ctr-interface.h @@ -0,0 +1,44 @@ +#ifndef _TIER_CTR_INTERFACE_H_ +#define _TIER_CTR_INTERFACE_H_ + +#include "common-utils.h" +#include "gfdb_data_store_types.h" + +#define GFDB_IPC_CTR_KEY "gfdb.ipc-ctr-op" + +/* + * CTR IPC OPERATIONS + * + * + */ +#define GFDB_IPC_CTR_QUERY_OPS "gfdb.ipc-ctr-query-op" +#define GFDB_IPC_CTR_CLEAR_OPS "gfdb.ipc-ctr-clear-op" +#define GFDB_IPC_CTR_GET_DB_PARAM_OPS "gfdb.ipc-ctr-get-db-parm" +#define GFDB_IPC_CTR_GET_DB_VERSION_OPS "gfdb.ipc-ctr-get-db-version" +#define GFDB_IPC_CTR_SET_COMPACT_PRAGMA "gfdb.ipc-ctr-set-compact-pragma" +/* + * CTR IPC INPUT/OUTPUT + * + * + */ +#define GFDB_IPC_CTR_GET_QFILE_PATH "gfdb.ipc-ctr-get-qfile-path" +#define GFDB_IPC_CTR_GET_QUERY_PARAMS "gfdb.ipc-ctr-get-query-parms" +#define GFDB_IPC_CTR_RET_QUERY_COUNT "gfdb.ipc-ctr-ret-rec-count" +#define GFDB_IPC_CTR_GET_DB_KEY "gfdb.ipc-ctr-get-params-key" +#define GFDB_IPC_CTR_RET_DB_VERSION "gfdb.ipc-ctr-ret-db-version" + +/* + * gfdb ipc ctr params for query + * + * + */ +typedef struct gfdb_ipc_ctr_params { + gf_boolean_t is_promote; + int write_freq_threshold; + int read_freq_threshold; + gfdb_time_t time_stamp; + int query_limit; + gf_boolean_t emergency_demote; +} gfdb_ipc_ctr_params_t; + +#endif -- cgit