election: Do not spam elections if one is already ongoing v1 APPLIED

Louis Solofrizzo: 1
 election: Do not spam elections if one is already ongoing

 3 files changed, 19 insertions(+), 5 deletions(-)

[PATCH] election: Do not spam elections if one is already ongoing Export this patch

And more debug here and there

Signed-off-by: Louis Solofrizzo <lsolofrizzo@scaleway.com>
---
 election.c | 19 ++++++++++++++++---
 libfloat.h |  1 +
 periodic.c |  4 ++--
 3 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/election.c b/election.c
index fbc15a3..c014756 100644
--- a/election.c
+++ b/election.c
@@ -24,6 +24,9 @@ void __libfloat_election_start(libfloat_ctx_t *ctx, libfloat_elections_args_t *a
     libfloat_term_t     last_term = 0;
     libfloat_entry_id_t last_id = 0;
 
+    if (ctx->state == RAFT_STATE_CANDIDATE && args->force == false)
+        return;
+
     DEBUG(ctx, "Election starting! reason=%s", args->reason);
     /* First, reset the vote of everyone */
     for_every_node(ctx, node, {
@@ -81,12 +84,15 @@ static bool libfloat_can_i_grant_vote(libfloat_ctx_t *ctx, libfloat_rpc_request_
     {
         if (ctx->persistent.voted_for == req->candidate_id)
             return true;
+
+        DEBUG(ctx, "Cannot grant vote to node %d: I have already voted for %d", req->candidate_id, ctx->persistent.voted_for);
         return false;
     }
 
     if (ctx->persistent.commit_index > 0 && libfloat_get_last_term(ctx, NULL, &last_term) == false)
     {
-        /* We have failed to retrieve last_ferm from log */
+        /* We have failed to retrieve last_term from log */
+        DEBUG(ctx, "Cannot grant vote to node %d: Cannot get last term", req->candidate_id);
         return false;
     }
 
@@ -94,6 +100,7 @@ static bool libfloat_can_i_grant_vote(libfloat_ctx_t *ctx, libfloat_rpc_request_
     if (last_term > req->last_log_term && req->last_log_index <= ctx->persistent.commit_index)
     {
         /* We have a superior term and a superior log, we can't grant our vote */
+        DEBUG(ctx, "Cannot grant vote to node %d: I have a greater term and a greater log", req->candidate_id);
         return false;
     }
 
@@ -103,10 +110,12 @@ static bool libfloat_can_i_grant_vote(libfloat_ctx_t *ctx, libfloat_rpc_request_
         /* We have a superior log, we can't grant our vote */
 
         /* There's an election in progress, and we have a superior log, let's try to become leader */
-        libfloat_election_start(ctx, .reason = "I have a greater log");
+        DEBUG(ctx, "Cannot grant vote to node %d: I have a greater log", req->candidate_id);
+        libfloat_election_start(ctx, .reason = "I have a greater log", .force = true);
         return false;
     }
 
+    DEBUG(ctx, "Granted vote to %d", req->candidate_id);
     /* All good! */
     return true;
 }
@@ -140,7 +149,11 @@ void libfloat_request_vote_receive(libfloat_ctx_t *ctx, libfloat_rpc_request_vot
     {
         libfloat_set_current_term(ctx, req->term);
 
-        libfloat_become_follower(ctx, .reason = "request_vote_receive: our term is lower than remote term");
+        if (ctx->state == RAFT_STATE_LEADER)
+            libfloat_become_follower(ctx, .reason = "request_vote_receive: our term is lower than remote term");
+
+        /* Vote reset */
+        libfloat_vote_for(ctx, 0);
         ctx->timeout_elapsed = 0;
     }
 
diff --git a/libfloat.h b/libfloat.h
index fe352d5..eb7056d 100644
--- a/libfloat.h
+++ b/libfloat.h
@@ -292,6 +292,7 @@ struct libfloat_ctx_s {
  */
 typedef struct {
     const char  *reason;        /*!< Logic reason for calling a function, for debug purposes */
+    bool        force;
 } libfloat_elections_args_t;
 
 /*!
diff --git a/periodic.c b/periodic.c
index cfab6cf..40004b4 100644
--- a/periodic.c
+++ b/periodic.c
@@ -93,7 +93,7 @@ void libfloat_periodic(libfloat_ctx_t *ctx, uint32_t time)
             {
                 /* DynamoDB-like elections are not implemented, let's simply start an election */
                 DEBUG(ctx, "New election: timeout elapsed %d", ctx->timeout_elapsed);
-                libfloat_election_start(ctx, .reason = "Election timeout, start a new one");
+                libfloat_election_start(ctx, .reason = "Election timeout, start a new one", .force = true);
                 ctx->lost_leader_time -= (ctx->conf.election_timeout / 1000);
             }
             else
@@ -162,7 +162,7 @@ void libfloat_periodic(libfloat_ctx_t *ctx, uint32_t time)
                         {
                             /* We did not receive any reponses, and the election timeout has expired twice, let's launch an election */
                             ctx->gray_failures.checking = false;
-                            libfloat_election_start(ctx, .reason = "Gray failures: Complete timeout");
+                            libfloat_election_start(ctx, .reason = "Gray failures: Complete timeout", .force = true);
 
                             /* XXX: In this specific case, I'm not sure that triggering an election actually does something:
                              * - Either every node in the cluster is down / unreachable, and a quorum will never be reached anyway
-- 
2.38.1