#include "lib/util/dlinklist.h"
#include "lib/util/debug.h"
#include "lib/util/samba_util.h"
+#include "lib/util/util_process.h"
#include "ctdb_private.h"
#include "ctdb_client.h"
-#include "ctdb_logging.h"
#include "common/rb_tree.h"
#include "common/reqid.h"
#include "common/system.h"
#include "common/common.h"
+#include "common/logging.h"
#define TAKEOVER_TIMEOUT() timeval_current_ofs(ctdb->tunable.takeover_timeout,0)
struct ctdb_ipflags {
bool noiptakeover;
bool noiphost;
- enum ctdb_runstate runstate;
+};
+
+enum ipalloc_algorithm {
+ IPALLOC_DETERMINISTIC,
+ IPALLOC_NONDETERMINISTIC,
+ IPALLOC_LCP2,
+};
+
+struct ipalloc_state {
+ uint32_t num;
+
+ /* Arrays with data for each node */
+ struct ctdb_public_ip_list_old **known_public_ips;
+ struct ctdb_public_ip_list_old **available_public_ips;
+
+ enum ipalloc_algorithm algorithm;
+ uint32_t no_ip_failback;
};
struct ctdb_interface {
{
int i;
+ /* Nodes that are not RUNNING can not host IPs */
+ if (ctdb->runstate != CTDB_RUNSTATE_RUNNING) {
+ return false;
+ }
+
if (vnn->delete_pending) {
return false;
}
/* Given a physical node, return the number of
public addresses that is currently assigned to this node.
*/
-static int node_ip_coverage(struct ctdb_context *ctdb, int32_t pnn,
- struct public_ip_list *ips)
+static int node_ip_coverage(int32_t pnn, struct public_ip_list *ips)
{
int num=0;
/* Can the given node host the given IP: is the public IP known to the
* node and is NOIPHOST unset?
*/
-static bool can_node_host_ip(struct ctdb_context *ctdb, int32_t pnn,
+static bool can_node_host_ip(struct ipalloc_state *ipalloc_state,
+ int32_t pnn,
struct ctdb_ipflags ipflags,
struct public_ip_list *ip)
{
return false;
}
- public_ips = ctdb->nodes[pnn]->available_public_ips;
+ public_ips = ipalloc_state->available_public_ips[pnn];
if (public_ips == NULL) {
return false;
return false;
}
-static bool can_node_takeover_ip(struct ctdb_context *ctdb, int32_t pnn,
+static bool can_node_takeover_ip(struct ipalloc_state *ipalloc_state,
+ int32_t pnn,
struct ctdb_ipflags ipflags,
struct public_ip_list *ip)
{
return false;
}
- return can_node_host_ip(ctdb, pnn, ipflags, ip);
+ return can_node_host_ip(ipalloc_state, pnn, ipflags, ip);
}
/* search the node lists list for a node to takeover this ip.
pick the node that currently are serving the least number of ips
so that the ips get spread out evenly.
*/
-static int find_takeover_node(struct ctdb_context *ctdb,
+static int find_takeover_node(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *ip,
struct public_ip_list *all_ips)
pnn = -1;
for (i=0; i<numnodes; i++) {
/* verify that this node can serve this ip */
- if (!can_node_takeover_ip(ctdb, i, ipflags[i], ip)) {
+ if (!can_node_takeover_ip(ipalloc_state, i, ipflags[i], ip)) {
/* no it couldnt so skip to the next node */
continue;
}
- num = node_ip_coverage(ctdb, i, all_ips);
+ num = node_ip_coverage(i, all_ips);
/* was this the first node we checked ? */
if (pnn == -1) {
pnn = i;
min = num;
}
}
- }
+ }
if (pnn == -1) {
DEBUG(DEBUG_WARNING,(__location__ " Could not find node to take over public address '%s'\n",
ctdb_addr_to_str(&ip->addr)));
return 0;
}
+static int verify_remote_ip_allocation(struct ctdb_context *ctdb,
+ struct ctdb_public_ip_list_old *ips,
+ uint32_t pnn);
+
+static int ctdb_reload_remote_public_ips(struct ctdb_context *ctdb,
+ struct ipalloc_state *ipalloc_state,
+ struct ctdb_node_map_old *nodemap)
+{
+ int j;
+ int ret;
+
+ if (ipalloc_state->num != nodemap->num) {
+ DEBUG(DEBUG_ERR,
+ (__location__
+ " ipalloc_state->num (%d) != nodemap->num (%d) invalid param\n",
+ ipalloc_state->num, nodemap->num));
+ return -1;
+ }
+
+ for (j=0; j<nodemap->num; j++) {
+ if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
+ continue;
+ }
+
+ /* Retrieve the list of known public IPs from the node */
+ ret = ctdb_ctrl_get_public_ips_flags(ctdb,
+ TAKEOVER_TIMEOUT(),
+ j,
+ ctdb->nodes,
+ 0,
+ &ipalloc_state->known_public_ips[j]);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR,
+ ("Failed to read known public IPs from node: %u\n",
+ j));
+ return -1;
+ }
+
+ if (ctdb->do_checkpublicip) {
+ verify_remote_ip_allocation(ctdb,
+ ipalloc_state->known_public_ips[j],
+ j);
+ }
+
+ /* Retrieve the list of available public IPs from the node */
+ ret = ctdb_ctrl_get_public_ips_flags(ctdb,
+ TAKEOVER_TIMEOUT(),
+ j,
+ ctdb->nodes,
+ CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE,
+ &ipalloc_state->available_public_ips[j]);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR,
+ ("Failed to read available public IPs from node: %u\n",
+ j));
+ return -1;
+ }
+ }
+
+ return 0;
+}
+
static struct public_ip_list *
-create_merged_ip_list(struct ctdb_context *ctdb)
+create_merged_ip_list(struct ctdb_context *ctdb, struct ipalloc_state *ipalloc_state)
{
int i, j;
struct public_ip_list *ip_list;
struct ctdb_public_ip_list_old *public_ips;
- if (ctdb->ip_tree != NULL) {
- talloc_free(ctdb->ip_tree);
- ctdb->ip_tree = NULL;
- }
+ TALLOC_FREE(ctdb->ip_tree);
ctdb->ip_tree = trbt_create(ctdb, 0);
- for (i=0;i<ctdb->num_nodes;i++) {
- public_ips = ctdb->nodes[i]->known_public_ips;
+ for (i=0; i < ctdb->num_nodes; i++) {
+ public_ips = ipalloc_state->known_public_ips[i];
if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
continue;
/* there were no public ips for this node */
if (public_ips == NULL) {
continue;
- }
+ }
- for (j=0;j<public_ips->num;j++) {
+ for (j=0; j < public_ips->num; j++) {
struct public_ip_list *tmp_ip;
tmp_ip = talloc_zero(ctdb->ip_tree, struct public_ip_list);
/* Allocate any unassigned IPs just by looping through the IPs and
* finding the best node for each.
*/
-static void basic_allocate_unassigned(struct ctdb_context *ctdb,
+static void basic_allocate_unassigned(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips)
{
struct public_ip_list *tmp_ip;
- /* loop over all ip's and find a physical node to cover for
+ /* loop over all ip's and find a physical node to cover for
each unassigned ip.
*/
for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
if (tmp_ip->pnn == -1) {
- if (find_takeover_node(ctdb, ipflags, tmp_ip, all_ips)) {
- DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
- ctdb_addr_to_str(&tmp_ip->addr)));
+ if (find_takeover_node(ipalloc_state, ipflags,
+ tmp_ip, all_ips)) {
+ DEBUG(DEBUG_WARNING,
+ ("Failed to find node to cover ip %s\n",
+ ctdb_addr_to_str(&tmp_ip->addr)));
}
}
}
/* Basic non-deterministic rebalancing algorithm.
*/
-static void basic_failback(struct ctdb_context *ctdb,
+static void basic_failback(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips,
int num_ips)
minnode = -1;
for (i=0; i<numnodes; i++) {
/* only check nodes that can actually serve this ip */
- if (!can_node_takeover_ip(ctdb, i, ipflags[i], tmp_ip)) {
+ if (!can_node_takeover_ip(ipalloc_state, i,
+ ipflags[i], tmp_ip)) {
/* no it couldnt so skip to the next node */
continue;
}
- num = node_ip_coverage(ctdb, i, all_ips);
+ num = node_ip_coverage(i, all_ips);
if (maxnode == -1) {
maxnode = i;
maxnum = num;
/* Reassign one of maxnode's VNNs */
for (tmp=all_ips;tmp;tmp=tmp->next) {
if (tmp->pnn == maxnode) {
- (void)find_takeover_node(ctdb, ipflags, tmp, all_ips);
+ (void)find_takeover_node(ipalloc_state,
+ ipflags,
+ tmp,
+ all_ips);
retries++;
goto try_again;;
}
}
}
-static void lcp2_init(struct ctdb_context *tmp_ctx,
+static void lcp2_init(TALLOC_CTX *tmp_ctx,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips,
uint32_t *force_rebalance_nodes,
/* Allocate any unassigned addresses using the LCP2 algorithm to find
* the IP/node combination that will cost the least.
*/
-static void lcp2_allocate_unassigned(struct ctdb_context *ctdb,
+static void lcp2_allocate_unassigned(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips,
uint32_t *lcp2_imbalances)
for (dstnode=0; dstnode<numnodes; dstnode++) {
/* only check nodes that can actually takeover this ip */
- if (!can_node_takeover_ip(ctdb, dstnode,
+ if (!can_node_takeover_ip(ipalloc_state,
+ dstnode,
ipflags[dstnode],
tmp_ip)) {
/* no it couldnt so skip to the next node */
* to move IPs from, determines the best IP/destination node
* combination to move from the source node.
*/
-static bool lcp2_failback_candidate(struct ctdb_context *ctdb,
+static bool lcp2_failback_candidate(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips,
int srcnode,
}
/* only check nodes that can actually takeover this ip */
- if (!can_node_takeover_ip(ctdb, dstnode,
+ if (!can_node_takeover_ip(ipalloc_state, dstnode,
ipflags[dstnode], tmp_ip)) {
/* no it couldnt so skip to the next node */
continue;
* node with the highest LCP2 imbalance, and then determines the best
* IP/destination node combination to move from the source node.
*/
-static void lcp2_failback(struct ctdb_context *ctdb,
+static void lcp2_failback(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips,
uint32_t *lcp2_imbalances,
*/
DEBUG(DEBUG_DEBUG,("+++++++++++++++++++++++++++++++++++++++++\n"));
DEBUG(DEBUG_DEBUG,("Selecting most imbalanced node from:\n"));
- lips = talloc_array(ctdb, struct lcp2_imbalance_pnn, numnodes);
+ lips = talloc_array(ipalloc_state, struct lcp2_imbalance_pnn, numnodes);
for (i=0; i<numnodes; i++) {
lips[i].imbalance = lcp2_imbalances[i];
lips[i].pnn = i;
break;
}
- if (lcp2_failback_candidate(ctdb,
+ if (lcp2_failback_candidate(ipalloc_state,
ipflags,
all_ips,
lips[i].pnn,
}
}
-static void unassign_unsuitable_ips(struct ctdb_context *ctdb,
+static void unassign_unsuitable_ips(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips)
{
if (tmp_ip->pnn == -1) {
continue;
}
- if (!can_node_host_ip(ctdb, tmp_ip->pnn,
+ if (!can_node_host_ip(ipalloc_state, tmp_ip->pnn,
ipflags[tmp_ip->pnn], tmp_ip) != 0) {
/* this node can not serve this ip. */
DEBUG(DEBUG_DEBUG,("Unassign IP: %s from %d\n",
}
}
-static void ip_alloc_deterministic_ips(struct ctdb_context *ctdb,
+static void ip_alloc_deterministic_ips(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips)
{
* IPs, since the modulo step above implicitly fails
* back IPs to their "home" node.
*/
- if (1 == ctdb->tunable.no_ip_failback) {
+ if (1 == ipalloc_state->no_ip_failback) {
DEBUG(DEBUG_WARNING, ("WARNING: 'NoIPFailback' set but ignored - incompatible with 'DeterministicIPs\n"));
}
- unassign_unsuitable_ips(ctdb, ipflags, all_ips);
+ unassign_unsuitable_ips(ipalloc_state, ipflags, all_ips);
- basic_allocate_unassigned(ctdb, ipflags, all_ips);
+ basic_allocate_unassigned(ipalloc_state, ipflags, all_ips);
/* No failback here! */
}
-static void ip_alloc_nondeterministic_ips(struct ctdb_context *ctdb,
+static void ip_alloc_nondeterministic_ips(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips)
{
num_ips++;
}
- unassign_unsuitable_ips(ctdb, ipflags, all_ips);
+ unassign_unsuitable_ips(ipalloc_state, ipflags, all_ips);
- basic_allocate_unassigned(ctdb, ipflags, all_ips);
+ basic_allocate_unassigned(ipalloc_state, ipflags, all_ips);
/* If we don't want IPs to fail back then don't rebalance IPs. */
- if (1 == ctdb->tunable.no_ip_failback) {
+ if (1 == ipalloc_state->no_ip_failback) {
return;
}
/* Now, try to make sure the ip adresses are evenly distributed
across the nodes.
*/
- basic_failback(ctdb, ipflags, all_ips, num_ips);
+ basic_failback(ipalloc_state, ipflags, all_ips, num_ips);
}
-static void ip_alloc_lcp2(struct ctdb_context *ctdb,
+static void ip_alloc_lcp2(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
struct public_ip_list *all_ips,
uint32_t *force_rebalance_nodes)
bool *rebalance_candidates;
int numnodes, num_rebalance_candidates, i;
- TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
+ TALLOC_CTX *tmp_ctx = talloc_new(ipalloc_state);
- unassign_unsuitable_ips(ctdb, ipflags, all_ips);
+ unassign_unsuitable_ips(ipalloc_state, ipflags, all_ips);
lcp2_init(tmp_ctx, ipflags, all_ips,force_rebalance_nodes,
&lcp2_imbalances, &rebalance_candidates);
- lcp2_allocate_unassigned(ctdb, ipflags, all_ips, lcp2_imbalances);
+ lcp2_allocate_unassigned(ipalloc_state, ipflags, all_ips, lcp2_imbalances);
/* If we don't want IPs to fail back then don't rebalance IPs. */
- if (1 == ctdb->tunable.no_ip_failback) {
+ if (1 == ipalloc_state->no_ip_failback) {
goto finished;
}
/* Now, try to make sure the ip adresses are evenly distributed
across the nodes.
*/
- lcp2_failback(ctdb, ipflags, all_ips,
+ lcp2_failback(ipalloc_state, ipflags, all_ips,
lcp2_imbalances, rebalance_candidates);
finished:
}
/* The calculation part of the IP allocation algorithm. */
-static void ctdb_takeover_run_core(struct ctdb_context *ctdb,
+static void ctdb_takeover_run_core(struct ipalloc_state *ipalloc_state,
struct ctdb_ipflags *ipflags,
- struct public_ip_list **all_ips_p,
+ struct public_ip_list *all_ips,
uint32_t *force_rebalance_nodes)
{
- /* since nodes only know about those public addresses that
- can be served by that particular node, no single node has
- 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_p = create_merged_ip_list(ctdb);
-
- if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
- ip_alloc_lcp2(ctdb, ipflags, *all_ips_p, force_rebalance_nodes);
- } else if (1 == ctdb->tunable.deterministic_public_ips) {
- ip_alloc_deterministic_ips(ctdb, ipflags, *all_ips_p);
- } else {
- ip_alloc_nondeterministic_ips(ctdb, ipflags, *all_ips_p);
+ switch (ipalloc_state->algorithm) {
+ case IPALLOC_LCP2:
+ ip_alloc_lcp2(ipalloc_state, ipflags, all_ips,
+ force_rebalance_nodes);
+ break;
+ case IPALLOC_DETERMINISTIC:
+ ip_alloc_deterministic_ips(ipalloc_state, ipflags, all_ips);
+ break;
+ case IPALLOC_NONDETERMINISTIC:
+ ip_alloc_nondeterministic_ips(ipalloc_state, ipflags, all_ips);
+ break;
}
/* at this point ->pnn is the node which will own each IP
return tvals;
}
-struct get_runstate_callback_data {
- enum ctdb_runstate *out;
- bool fatal;
-};
-
-static void get_runstate_callback(struct ctdb_context *ctdb, uint32_t pnn,
- int32_t res, TDB_DATA outdata,
- void *callback_data)
-{
- struct get_runstate_callback_data *cd =
- (struct get_runstate_callback_data *)callback_data;
- int size;
-
- if (res != 0) {
- /* Already handled in fail callback */
- return;
- }
-
- if (outdata.dsize != sizeof(uint32_t)) {
- DEBUG(DEBUG_ERR,("Wrong size of returned data when getting runstate from node %d. Expected %d bytes but received %d bytes\n",
- pnn, (int)sizeof(uint32_t),
- (int)outdata.dsize));
- cd->fatal = true;
- return;
- }
-
- size = talloc_array_length(cd->out);
- if (pnn >= size) {
- DEBUG(DEBUG_ERR,("Got reply from node %d but nodemap only has %d entries\n",
- pnn, size));
- return;
- }
-
- cd->out[pnn] = (enum ctdb_runstate)*(uint32_t *)outdata.dptr;
-}
-
-static void get_runstate_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
- int32_t res, TDB_DATA outdata,
- void *callback)
-{
- struct get_runstate_callback_data *cd =
- (struct get_runstate_callback_data *)callback;
-
- switch (res) {
- case -ETIME:
- DEBUG(DEBUG_ERR,
- ("Timed out getting runstate from node %d\n", pnn));
- cd->fatal = true;
- break;
- default:
- DEBUG(DEBUG_WARNING,
- ("Error getting runstate from node %d - assuming runstates not supported\n",
- pnn));
- }
-}
-
-static enum ctdb_runstate * get_runstate_from_nodes(struct ctdb_context *ctdb,
- TALLOC_CTX *tmp_ctx,
- struct ctdb_node_map_old *nodemap,
- enum ctdb_runstate default_value)
-{
- uint32_t *nodes;
- enum ctdb_runstate *rs;
- struct get_runstate_callback_data callback_data;
- int i;
-
- rs = talloc_array(tmp_ctx, enum ctdb_runstate, nodemap->num);
- CTDB_NO_MEMORY_NULL(ctdb, rs);
- for (i=0; i<nodemap->num; i++) {
- rs[i] = default_value;
- }
-
- callback_data.out = rs;
- callback_data.fatal = false;
-
- nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
- if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_RUNSTATE,
- nodes, 0, TAKEOVER_TIMEOUT(),
- true, tdb_null,
- get_runstate_callback,
- get_runstate_fail_callback,
- &callback_data) != 0) {
- if (callback_data.fatal) {
- free(rs);
- rs = NULL;
- }
- }
- talloc_free(nodes);
-
- return rs;
-}
-
/* Set internal flags for IP allocation:
* Clear ip flags
* Set NOIPTAKOVER ip flags from per-node NoIPTakeover tunable
TALLOC_CTX *tmp_ctx,
struct ctdb_node_map_old *nodemap,
uint32_t *tval_noiptakeover,
- uint32_t *tval_noiphostonalldisabled,
- enum ctdb_runstate *runstate)
+ uint32_t *tval_noiphostonalldisabled)
{
int i;
struct ctdb_ipflags *ipflags;
ipflags[i].noiptakeover = true;
}
- /* Can not host IPs on node not in RUNNING state */
- if (runstate[i] != CTDB_RUNSTATE_RUNNING) {
- ipflags[i].noiphost = true;
- continue;
- }
/* Can not host IPs on INACTIVE node */
if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
ipflags[i].noiphost = true;
}
- /* Remember the runstate */
- ipflags[i].runstate = runstate[i];
}
if (all_nodes_are_disabled(nodemap)) {
uint32_t *tval_noiptakeover;
uint32_t *tval_noiphostonalldisabled;
struct ctdb_ipflags *ipflags;
- enum ctdb_runstate *runstate;
tval_noiptakeover = get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
return NULL;
}
- /* Any nodes where CTDB_CONTROL_GET_RUNSTATE is not supported
- * will default to CTDB_RUNSTATE_RUNNING. This ensures
- * reasonable behaviour on a mixed cluster during upgrade.
- */
- runstate = get_runstate_from_nodes(ctdb, tmp_ctx, nodemap,
- CTDB_RUNSTATE_RUNNING);
- if (runstate == NULL) {
- /* Caller frees tmp_ctx */
- return NULL;
- }
-
ipflags = set_ipflags_internal(ctdb, tmp_ctx, nodemap,
tval_noiptakeover,
- tval_noiphostonalldisabled,
- runstate);
+ tval_noiphostonalldisabled);
talloc_free(tval_noiptakeover);
talloc_free(tval_noiphostonalldisabled);
- talloc_free(runstate);
return ipflags;
}
+static struct ipalloc_state * ipalloc_state_init(struct ctdb_context *ctdb,
+ TALLOC_CTX *mem_ctx)
+{
+ struct ipalloc_state *ipalloc_state =
+ talloc_zero(mem_ctx, struct ipalloc_state);
+ if (ipalloc_state == NULL) {
+ DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
+ return NULL;
+ }
+
+ ipalloc_state->num = ctdb->num_nodes;
+ ipalloc_state->known_public_ips =
+ talloc_zero_array(ipalloc_state,
+ struct ctdb_public_ip_list_old *,
+ ipalloc_state->num);
+ if (ipalloc_state->known_public_ips == NULL) {
+ DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
+ talloc_free(ipalloc_state);
+ return NULL;
+ }
+ ipalloc_state->available_public_ips =
+ talloc_zero_array(ipalloc_state,
+ struct ctdb_public_ip_list_old *,
+ ipalloc_state->num);
+ if (ipalloc_state->available_public_ips == NULL) {
+ DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
+ talloc_free(ipalloc_state);
+ return NULL;
+ }
+
+ if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
+ ipalloc_state->algorithm = IPALLOC_LCP2;
+ } else if (1 == ctdb->tunable.deterministic_public_ips) {
+ ipalloc_state->algorithm = IPALLOC_DETERMINISTIC;
+ } else {
+ ipalloc_state->algorithm = IPALLOC_NONDETERMINISTIC;
+ }
+
+ ipalloc_state->no_ip_failback = ctdb->tunable.no_ip_failback;
+
+ return ipalloc_state;
+}
+
struct iprealloc_callback_data {
bool *retry_nodes;
int retry_count;
struct ctdb_client_control_state *state;
TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
struct ctdb_ipflags *ipflags;
+ struct ipalloc_state *ipalloc_state;
struct takeover_callback_data *takeover_data;
struct iprealloc_callback_data iprealloc_data;
bool *retry_data;
goto ipreallocated;
}
+ ipalloc_state = ipalloc_state_init(ctdb, tmp_ctx);
+ if (ipalloc_state == NULL) {
+ talloc_free(tmp_ctx);
+ return -1;
+ }
+
ipflags = set_ipflags(ctdb, tmp_ctx, nodemap);
if (ipflags == NULL) {
DEBUG(DEBUG_ERR,("Failed to set IP flags - aborting takeover run\n"));
return -1;
}
- /* Short-circuit IP allocation if no nodes are in the RUNNING
- * runstate yet, since no nodes will be able to host IPs */
+ /* Fetch known/available public IPs from each active node */
+ ret = ctdb_reload_remote_public_ips(ctdb, ipalloc_state, nodemap);
+ if (ret != 0) {
+ talloc_free(tmp_ctx);
+ return -1;
+ }
+
+ /* Short-circuit IP allocation if no node has available IPs */
can_host_ips = false;
- for (i=0; i<nodemap->num; i++) {
- if (ipflags[i].runstate == CTDB_RUNSTATE_RUNNING) {
+ for (i=0; i < ipalloc_state->num; i++) {
+ if (ipalloc_state->available_public_ips[i] != NULL) {
can_host_ips = true;
}
}
return 0;
}
+ /* since nodes only know about those public addresses that
+ can be served by that particular node, no single node has
+ 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, ipalloc_state);
+
/* Do the IP reassignment calculations */
- ctdb_takeover_run_core(ctdb, ipflags, &all_ips, force_rebalance_nodes);
+ ctdb_takeover_run_core(ipalloc_state, ipflags,
+ all_ips, force_rebalance_nodes);
/* Now tell all nodes to release any public IPs should not
* host. This will be a NOOP on nodes that don't currently
node has the expected ip allocation.
This is verified against ctdb->ip_tree
*/
-int verify_remote_ip_allocation(struct ctdb_context *ctdb,
- struct ctdb_public_ip_list_old *ips,
- uint32_t pnn)
+static int verify_remote_ip_allocation(struct ctdb_context *ctdb,
+ struct ctdb_public_ip_list_old *ips,
+ uint32_t pnn)
{
struct public_ip_list *tmp_ip;
int i;
close(h->fd[0]);
debug_extra = talloc_asprintf(NULL, "reloadips:");
- ctdb_set_process_name("ctdb_reloadips");
+ prctl_set_comment("ctdb_reloadips");
if (switch_from_server_to_client(ctdb, "reloadips-child") != 0) {
DEBUG(DEBUG_CRIT,("ERROR: Failed to switch reloadips child into client mode\n"));
res = -1;
git.samba.org / obnox / samba / samba-obnox.git / blobdiff