Akmal, When you say virtual disk are they backed by a local hdd per node or backed by shared storage over network.
I have seen this issue before due to network disruption/load and/or due to shared storage unable to keep up. Seems like increasing fsync limit is a favorite option among folks who encounter this. I would recommend addressing the root cause. Since ZK is a distributed system there is no need for use of shared storage in my opinion with regards to functional aspect of any deployment. Can you let us know your requirement for ZK deployment in Azure ?. Thanks Powell Sent from my iPhone > On Dec 18, 2015, at 12:19 AM, Akmal Abbasov <[email protected]> wrote: > > No, I’m using separate virtual disks to keep logs and data. But still I have > very high number of timeouts due to fsync, actually followers are timing and > the leader is closing connection to them. > Which starts new round of elections. This happens very frequently. There are > 3 servers in ensemble. > Settings I’m using > ticktime = 3000 > syncLimit=2 > …. > Now I can see only 3 possible solutions: > 1. Increase syncTime > 2. Set forceSync to no > 3. Update storage from hdd to sdd. > > Any thoughts about pros and cons of each solution? > Thank you. > > Regards, Akmal > >> On 17 Dec 2015, at 16:52, Flavio Junqueira <[email protected]> wrote: >> >> Akmal, >> >> What are you using for storage, local temporary storage? How many servers do >> you have in your ensemble? Do you have a virtual net setup? The throughput >> does seem pretty low. >> >> -Flavio >> >>> On 17 Dec 2015, at 15:27, Akmal Abbasov <[email protected]> wrote: >>> >>> Hi, >>> I was wondering does anybody have an experience running Zookeeper in Azure. >>> The current load is around 50 writes per second. Which type of storage >>> should be used, will basic work or SSD should be used instead? >>> Thank you. >>> >>> Regards, Akmal >
