From 89d31da66ca3f15ed3a87851dea5422be949a2e4 Mon Sep 17 00:00:00 2001 From: Ronnie Sahlberg Date: Thu, 8 Apr 2010 14:07:57 +1000 Subject: [PATCH] In the recovery daemon, keep track of which node we have assigned public ip addresses and verify that the remote nodes have/keep a consistent view of assigned addresses. If a remote node has an inconsistent view of addresses visavi the recovery master this will trigger a full ip reallocation. --- common/rb_tree.h | 12 ++++----- include/ctdb_private.h | 8 +++++- server/ctdb_recoverd.c | 11 +++++--- server/ctdb_takeover.c | 57 ++++++++++++++++++++++++++++++++++++------ 4 files changed, 71 insertions(+), 17 deletions(-) diff --git a/common/rb_tree.h b/common/rb_tree.h index cb7cba3b..eef0bc52 100644 --- a/common/rb_tree.h +++ b/common/rb_tree.h @@ -21,17 +21,17 @@ #define TRBT_RED 0x00 #define TRBT_BLACK 0x01 -typedef struct _trbt_node_t { - struct _trbt_tree_t *tree; - struct _trbt_node_t *parent; - struct _trbt_node_t *left; - struct _trbt_node_t *right; +typedef struct trbt_node { + struct trbt_tree *tree; + struct trbt_node *parent; + struct trbt_node *left; + struct trbt_node *right; uint32_t rb_color; uint32_t key32; void *data; } trbt_node_t; -typedef struct _trbt_tree_t { +typedef struct trbt_tree { trbt_node_t *root; /* automatically free the tree when the last node has been deleted */ #define TRBT_AUTOFREE 0x00000001 diff --git a/include/ctdb_private.h b/include/ctdb_private.h index f5e6de8e..06985a3c 100644 --- a/include/ctdb_private.h +++ b/include/ctdb_private.h @@ -439,7 +439,7 @@ struct ctdb_context { struct ctdb_call_state *pending_calls; struct ctdb_client_ip *client_ip_list; bool do_checkpublicip; - struct _trbt_tree_t *server_ids; + struct trbt_tree *server_ids; const char *event_script_dir; const char *notification_script; const char *default_public_interface; @@ -466,6 +466,9 @@ struct ctdb_context { /* mapping from pid to ctdb_client * */ struct ctdb_client_pid_list *client_pids; + + /* used in the recovery daemon to remember the ip allocation */ + struct trbt_tree *ip_tree; }; struct ctdb_db_context { @@ -1561,4 +1564,7 @@ int ctdb_update_persistent_health(struct ctdb_context *ctdb, int num_healthy_nodes); int ctdb_recheck_persistent_health(struct ctdb_context *ctdb); +int verify_remote_ip_allocation(struct ctdb_context *ctdb, + struct ctdb_all_public_ips *ips); + #endif diff --git a/server/ctdb_recoverd.c b/server/ctdb_recoverd.c index f94d43bd..68f744af 100644 --- a/server/ctdb_recoverd.c +++ b/server/ctdb_recoverd.c @@ -2321,9 +2321,9 @@ static enum monitor_result verify_recmaster(struct ctdb_recoverd *rec, struct ct } -/* called to check that the allocation of public ip addresses is ok. +/* called to check that the local allocation of public ip addresses is ok. */ -static int verify_ip_allocation(struct ctdb_context *ctdb, struct ctdb_recoverd *rec, uint32_t pnn) +static int verify_local_ip_allocation(struct ctdb_context *ctdb, struct ctdb_recoverd *rec, uint32_t pnn) { TALLOC_CTX *mem_ctx = talloc_new(NULL); struct ctdb_all_public_ips *ips = NULL; @@ -2973,7 +2973,7 @@ again: */ if (ctdb->do_checkpublicip) { if (rec->ip_check_disable_ctx == NULL) { - if (verify_ip_allocation(ctdb, rec, pnn) != 0) { + if (verify_local_ip_allocation(ctdb, rec, pnn) != 0) { DEBUG(DEBUG_ERR, (__location__ " Public IPs were inconsistent.\n")); } } @@ -3028,6 +3028,11 @@ again: ctdb->nodes[j]->pnn)); goto again; } + + if (verify_remote_ip_allocation(ctdb, ctdb->nodes[j]->public_ips)) { + DEBUG(DEBUG_ERR,("Node %d has inconsistent public ip allocation and needs update.\n", ctdb->nodes[j]->pnn)); + rec->need_takeover_run = true; + } } diff --git a/server/ctdb_takeover.c b/server/ctdb_takeover.c index d685de8d..acc9ce30 100644 --- a/server/ctdb_takeover.c +++ b/server/ctdb_takeover.c @@ -686,14 +686,17 @@ void getips_count_callback(void *param, void *data) } struct ctdb_public_ip_list * -create_merged_ip_list(struct ctdb_context *ctdb, TALLOC_CTX *tmp_ctx) +create_merged_ip_list(struct ctdb_context *ctdb) { int i, j; struct ctdb_public_ip_list *ip_list; struct ctdb_all_public_ips *public_ips; - trbt_tree_t *ip_tree; - ip_tree = trbt_create(tmp_ctx, 0); + if (ctdb->ip_tree != NULL) { + talloc_free(ctdb->ip_tree); + ctdb->ip_tree = NULL; + } + ctdb->ip_tree = trbt_create(ctdb, 0); for (i=0;inum_nodes;i++) { public_ips = ctdb->nodes[i]->public_ips; @@ -710,13 +713,13 @@ create_merged_ip_list(struct ctdb_context *ctdb, TALLOC_CTX *tmp_ctx) for (j=0;jnum;j++) { struct ctdb_public_ip_list *tmp_ip; - tmp_ip = talloc_zero(tmp_ctx, struct ctdb_public_ip_list); + tmp_ip = talloc_zero(ctdb->ip_tree, struct ctdb_public_ip_list); CTDB_NO_MEMORY_NULL(ctdb, tmp_ip); tmp_ip->pnn = public_ips->ips[j].pnn; tmp_ip->addr = public_ips->ips[j].addr; tmp_ip->next = NULL; - trbt_insertarray32_callback(ip_tree, + trbt_insertarray32_callback(ctdb->ip_tree, IP_KEYLEN, ip_key(&public_ips->ips[j].addr), add_ip_callback, tmp_ip); @@ -724,7 +727,7 @@ create_merged_ip_list(struct ctdb_context *ctdb, TALLOC_CTX *tmp_ctx) } ip_list = NULL; - trbt_traversearray32(ip_tree, IP_KEYLEN, getips_count_callback, &ip_list); + trbt_traversearray32(ctdb->ip_tree, IP_KEYLEN, getips_count_callback, &ip_list); return ip_list; } @@ -774,8 +777,10 @@ int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap) a full list of all public addresses that exist in the cluster. Walk over all node structures and create a merged list of all public addresses that exist in the cluster. + + keep the tree of ips around as ctdb->ip_tree */ - all_ips = create_merged_ip_list(ctdb, tmp_ctx); + all_ips = create_merged_ip_list(ctdb); /* If we want deterministic ip allocations, i.e. that the ip addresses will always be allocated the same way for a specific set of @@ -2149,3 +2154,41 @@ int32_t ctdb_control_del_public_address(struct ctdb_context *ctdb, TDB_DATA inda return -1; } +/* This function is called from the recovery daemon to verify that a remote + node has the expected ip allocation. + This is verified against ctdb->ip_tree +*/ +int verify_remote_ip_allocation(struct ctdb_context *ctdb, struct ctdb_all_public_ips *ips) +{ + struct ctdb_public_ip_list *tmp_ip; + int i; + + if (ctdb->ip_tree == NULL) { + /* dont know the expected allocation yet, assume remote node + is correct. */ + return 0; + } + + if (ips == NULL) { + return 0; + } + + for (i=0; inum; i++) { + tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ips->ips[i].addr)); + if (tmp_ip == NULL) { + DEBUG(DEBUG_ERR,(__location__ " Could not find host for address %s, reassign ips\n", ctdb_addr_to_str(&ips->ips[i].addr))); + return -1; + } + + if (tmp_ip->pnn == -1 || ips->ips[i].pnn == -1) { + continue; + } + + if (tmp_ip->pnn != ips->ips[i].pnn) { + DEBUG(DEBUG_ERR,("Inconsistent ip allocation. Trigger reallocation.\n")); + return -1; + } + } + + return 0; +} -- 2.34.1