реплика шарда будет лежать в двух разных ЦОДах. Кто-нибудь уже так делал? Какие проблемы возможны? Задержка на реплицирование?
Алексей делал https://clickhouse.com/blog/building-a-paste-service-with-clickhouse
у меня по европе раскидано все германия несколько локаций , хельсинки , амстердам и еще какая то , в целом задержка приемлемая
We have a cluster with two replicas in Europe and one replica in the US. ZooKeeper is a 3-node cluster, it is located in Europe in the same DC with one of the replicas and we do all INSERTs in that DC (directly into Replicated* tables). The US replica is configured with replicated_can_become_leader=0 (I don't remember the details, but we did have issues when the US replica acquired leadership on some tables). We tried at first to set up a couple of additional ZooKeeper nodes in the US, but then one of the European nodes went down and the ZooKeeper cluster was waiting for one of the US nodes to commit every transaction and it was extremely bad, inserts were running incredibly slow. When there is no connectivity issues the US replica works very well. There are data delays, but we don't notice them much (only when there are connectivity issues). The US replica restarts 5 times longer than the European replicas though, probably because we have about 200 replicated tables. We are thinking of installing a ZooKeeper observer node in the US, but we haven't done it yet (I guess it means that it's not really necessary 🙂 )
Обсуждают сегодня