Ack.
However, the patch's slogan is a bit unclear to me. ;)
Thanks, Vu
On 4/24/20 3:31 PM, thang.d.nguyen wrote:
When NFS is unavailable, client try to write to log. Lgs server
will put it into the queue with the time. This info will check point
with standby. Switch-over happens, and NFS is available again. New active
will get the diffent time b/w current time and the time put into the queue.
The timer on node can be different and it causes the coredump due to
the current time less then time put into the queue.
Once node is active, it must update the time put into the queue to
make it consistent on local node.
---
src/log/logd/lgs_cache.cc | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/src/log/logd/lgs_cache.cc b/src/log/logd/lgs_cache.cc
index d6a282e48..e3583e97c 100644
--- a/src/log/logd/lgs_cache.cc
+++ b/src/log/logd/lgs_cache.cc
@@ -124,7 +124,11 @@ Cache::Data::Data(std::shared_ptr<WriteAsyncInfo> info,
Cache::Data::Data(const CkptPushAsync* data) {
TRACE_ENTER();
param_ = std::make_shared<WriteAsyncInfo>(data);
- queue_at_ = data->queue_at;
+ // Don't inherit the queue_at_ from the active node,
+ // since the timer on both nodes may be different.
+ // Queue_at now is about when the element is actually
+ // put into the queue of each logsv instance.
+ queue_at_ = base::TimespecToNanos(base::ReadMonotonicClock());
seq_id_ = data->seq_id;
log_record_ = strdup(data->log_record);
size_ = strlen(log_record_);
_______________________________________________
Opensaf-devel mailing list
Opensaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opensaf-devel
