Check qorum is ok in periodic check v1 APPLIED

Michael Bonfils: 1
 Check qorum is ok in periodic check

 1 files changed, 10 insertions(+), 6 deletions(-)

[PATCH] Check qorum is ok in periodic check Export this patch

Stopped followers didn't trigger a step down from leader
if qorum is no more reached.

Signed-off-by: Michael Bonfils <mbonfils@scaleway.com>
---
 periodic.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/periodic.c b/periodic.c
index 94bf728..f3a95e6 100644
--- a/periodic.c
+++ b/periodic.c
@@ -10,25 +10,29 @@ void libfloat_periodic(libfloat_ctx_t *ctx, uint32_t time)
         {
             if (ctx->conf.sanity_timeout != 0 && ctx->n_nodes > 1)
             {
-                bool            node_reachable = false;
+                int             nodes_reachables = 0;
                 libfloat_node_t *node;
 
                 for_every_node(ctx, node, {
-                    if (node->last_update + ctx->conf.sanity_timeout > ctx->time(NULL))
+                    if (node->id == ctx->me->id)
                     {
-                        node_reachable = true;
-                        break;
+                        nodes_reachables ++;
+                    }
+                    else if (node->last_update + ctx->conf.sanity_timeout > ctx->time(NULL))
+                    {
+                        nodes_reachables ++;

Louis Solofrizzo <louis@ne02ptzero.me>

9 months ago

node_reachables++;

LGTM
-- 
Louis Solofrizzo
Team Storage
> "jackpot: you may have an unnecessary change record"
> -- message from "diff"

                     }
                 });
 
-                if (!node_reachable)
+                if (nodes_reachables < ctx->n_nodes / 2 + 1)
                 {
                     /**
                     * Hmmm, we might be partitionned, or at the very least we don't have a quorum anymore.
                     * Time to step down!
                     */
                     libfloat_become_follower(ctx);
-                    ERROR(ctx, "Sanity timeout has been reached (%d seconds), stepping down from leader position", ctx->conf.sanity_timeout);
+                    ERROR(ctx, "Sanity timeout has been reached (%d seconds), stepping down from leader position: reachable %d / quorum %lu (%lu)", 
+                          ctx->conf.sanity_timeout, nodes_reachables, ctx->n_nodes / 2 + 1, ctx->n_nodes);
                     return;
                 }
             }
-- 
2.25.1

Patrik Cyvoct <patrik@ptrk.io>

9 months ago

LG