mesos-master crash with zookeeper cluster - apache-zookeeper

Question

I am deploying a zookeeper cluster which has 3 nodes. I use it to keep my mesos master high availability. I download the zookeeper-3.4.6.tar.gz tarball and uncompress it to /opt, rename it to /opt/zookeeper, enter the directory, edit the conf/zoo.cfg(pasted below), create a myid file in dataDir(which is set to /var/lib/zookeeper in zoo.cfg), and start zookeeper using ./bin/zkServer.sh start, and it goes well. I start all the 3 nodes one by one and they all seems well. I use ./bin/zkCli.sh to connect the server , no problem.
But when I start mesos (3 masters and 3 slaves, each node runs a master and a slave), then the masters soon crashed, one by one, and in the webpage http://mesos_master:5050, slave tab, no slaves are displayed. But when I run only one zookeeper, these are all fine. So I think it's the zookeeper cluster's problem.
I got 3 PV host in my ubuntu server. they are all running ubuntu 14.04 LTS:
node-01, node-02, node-03,
I have /etc/hosts in all three nodes like this:
172.16.2.70 node-01
172.16.2.81 node-02
172.16.2.80 node-03
I installed zookeeper, mesos on all the three nodes. Zookeeper configure file is like this (all three nodes) :
tickTime=2000
dataDir=/var/lib/zookeeper
clientPort=2181
initLimit=5
syncLimit=2
server.1=node-01:2888:3888
server.2=node-02:2888:3888
server.3=node-03:2888:3888
they can be started normally and run well. And then I start the mesos-master service, using the command line ./bin/mesos-master.sh --zk=zk://172.16.2.70:2181,172.16.2.81:2181,172.16.2.80:2181/mesos --work_dir=/var/lib/mesos --quorum=2, and after a few seconds, it gives me errors like this:
F0817 15:09:19.995256 2250 master.cpp:1253] Recovery failed: Failed to recover registrar: Failed to perform fetch within 1mins
*** Check failure stack trace: ***
# 0x7fa2b8be71a2 google::LogMessage::Fail()
# 0x7fa2b8be70ee google::LogMessage::SendToLog()
# 0x7fa2b8be6af0 google::LogMessage::Flush()
# 0x7fa2b8be9a04 google::LogMessageFatal::~LogMessageFatal()
▽
# 0x7fa2b81a899a mesos::internal::master::fail()
▽
# 0x7fa2b8262f8f _ZNSt5_BindIFPFvRKSsS1_EPKcSt12_PlaceholderILi1EEEE6__callIvJS1_EJLm0ELm1EEEET_OSt5tupleIJDpT0_EESt12_Index_tupleIJXspT1_EEE
▽
# 0x7fa2b823fba7 _ZNSt5_BindIFPFvRKSsS1_EPKcSt12_PlaceholderILi1EEEEclIJS1_EvEET0_DpOT_
# 0x7fa2b820f9f3 _ZZNK7process6FutureI7NothingE8onFailedISt5_BindIFPFvRKSsS6_EPKcSt12_PlaceholderILi1EEEEvEERKS2_OT_NS2_6PreferEENUlS6_E_clES6_
# 0x7fa2b826305c _ZNSt17_Function_handlerIFvRKSsEZNK7process6FutureI7NothingE8onFailedISt5_BindIFPFvS1_S1_EPKcSt12_PlaceholderILi1EEEEvEERKS6_OT_NS6_6PreferEEUlS1_E_E9_M_invokeERKSt9_Any_dataS1_
# 0x4a44e7 std::function<>::operator()()
# 0x49f3a7 _ZN7process8internal3runISt8functionIFvRKSsEEJS4_EEEvRKSt6vectorIT_SaIS8_EEDpOT0_
# 0x499480 process::Future<>::fail()
# 0x7fa2b806b4b4 process::Promise<>::fail()
# 0x7fa2b826011b process::internal::thenf<>()
# 0x7fa2b82a0757 _ZNSt5_BindIFPFvRKSt8functionIFN7process6FutureI7NothingEERKN5mesos8internal8RegistryEEERKSt10shared_ptrINS1_7PromiseIS3_EEERKNS2_IS7_EEESB_SH_St12_PlaceholderILi1EEEE6__callIvISM_EILm0ELm1ELm2EEEET_OSt5tupleIIDpT0_EESt12_Index_tupleIIXspT1_EEE
# 0x7fa2b82962d9 std::_Bind<>::operator()<>()
# 0x7fa2b827ee89 std::_Function_handler<>::_M_invoke()
I0817 15:09:20.098639 2248 http.cpp:283] HTTP GET for /master/state.json from 172.16.2.84:54542 with User-Agent='Mozilla/5.0 (Macintosh; Intel Mac OS X 10_10_4) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/44.0.2403.155 Safari/537.36'
# 0x7fa2b8296507 std::function<>::operator()()
# 0x7fa2b827efaf _ZZNK7process6FutureIN5mesos8internal8RegistryEE5onAnyIRSt8functionIFvRKS4_EEvEES8_OT_NS4_6PreferEENUlS8_E_clES8_
# 0x7fa2b82a07fe _ZNSt17_Function_handlerIFvRKN7process6FutureIN5mesos8internal8RegistryEEEEZNKS5_5onAnyIRSt8functionIS8_EvEES7_OT_NS5_6PreferEEUlS7_E_E9_M_invokeERKSt9_Any_dataS7_
# 0x7fa2b8296507 std::function<>::operator()()
# 0x7fa2b82e4419 process::internal::run<>()
# 0x7fa2b82da22a process::Future<>::fail()
# 0x7fa2b83136b5 std::_Mem_fn<>::operator()<>()
# 0x7fa2b830efdf _ZNSt5_BindIFSt7_Mem_fnIMN7process6FutureIN5mesos8internal8RegistryEEEFbRKSsEES6_St12_PlaceholderILi1EEEE6__callIbIS8_EILm0ELm1EEEET_OSt5tupleIIDpT0_EESt12_Index_tupleIIXspT1_EEE
# 0x7fa2b8307d7f _ZNSt5_BindIFSt7_Mem_fnIMN7process6FutureIN5mesos8internal8RegistryEEEFbRKSsEES6_St12_PlaceholderILi1EEEEclIJS8_EbEET0_DpOT_
# 0x7fa2b82fe431 _ZZNK7process6FutureIN5mesos8internal8RegistryEE8onFailedISt5_BindIFSt7_Mem_fnIMS4_FbRKSsEES4_St12_PlaceholderILi1EEEEbEERKS4_OT_NS4_6PreferEENUlS9_E_clES9_
# 0x7fa2b830f065 _ZNSt17_Function_handlerIFvRKSsEZNK7process6FutureIN5mesos8internal8RegistryEE8onFailedISt5_BindIFSt7_Mem_fnIMS8_FbS1_EES8_St12_PlaceholderILi1EEEEbEERKS8_OT_NS8_6PreferEEUlS1_E_E9_M_invokeERKSt9_Any_dataS1_
# 0x4a44e7 std::function<>::operator()()
# 0x49f3a7 _ZN7process8internal3runISt8functionIFvRKSsEEJS4_EEEvRKSt6vectorIT_SaIS8_EEDpOT0_
# 0x7fa2b82da202 process::Future<>::fail()
# 0x7fa2b82d2d82 process::Promise<>::fail()
Aborted
sometimes the warning is like this, and then crashed with the same output above:
0817 15:09:49.745750 2104 recover.cpp:111] Unable to finish the recover protocol in 10secs, retrying
I want to know whether zookeeper is deployed and run well in my case, and How can I locate where the problem is. Any answers and suggests are welcomed. thanks.

Answer 1

Actually, in my case, It's because I didn't open firewall port 5050 to allow three servers to communicate with each others. After updating firewall rule, it starts to work as expected.

Answer 2

I fall into same issue, I tried different ways and different options and finally --ip option worked for me. Initially I used --hostname option
mesos-master --ip=192.168.0.13 --quorum=2 --zk=zk://m1:2181,m2:2181,m3:2181/mesos --work_dir=/opt/mm1 --log_dir=/opt/mm1/logs

Answer 3

You need to check that all mesos/zookeeper master nodes can communicate correctly. For that, you need:
Zookeeper ports open: TCP 2181, 2888, 3888
Mesos port open: TCP 5050
ping available (ICMP message 0 and 8)
If you use FQDN instead of IP in your config, check that the DNS resolution is working correctly as well.

Answer 4

Split your mesos masters' work_dir to different dir, do not use a share work_dir for all masters, because of zk