fix: retry starting zookeeper if it failed on onebox environment

[neverchanje]

What problem does this PR solve?

run.sh test doesn't check for availability of zookeeper when it fails to start, it will continue and fail eventually when some unit tests (function_test) detect that the onebox is not started.

~/build/XiaoMi/pegasus/src/builder/bin/pegasus_function_test ~/build/XiaoMi/pegasus
============
start global_env()
meta1 pid: 0
pegasus_function_test: /home/travis/build/XiaoMi/pegasus/src/test/function_test/global_env.cpp:39: void global_env::get_dirs(): Assertion `dsn::utils::pipe_execute(cmd2, ss2) == 0' failed.
./run.sh: line 35: 18247 Aborted                 (core dumped) GTEST_OUTPUT="xml:$REPORT_DIR/basic.xml" GTEST_FILTER="basic.*" ./$test_case $config_file $table_name

Decompressing zookeeper...
JMX enabled by default
Using config: /home/travis/build/XiaoMi/pegasus/.zk_install/zookeeper-3.4.6/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
ERROR: start zookeeper failed

Actually the zookeeper may not be failed at first, it just needs a little bit longer to wait until fully bootstrapped.

What is changed and how it works?

In this PR we check if zookeeper is started failed, if true, the script fails fast. If the onebox fails, the script fails fast. We wait 30 seconds for the bootstrap of zookeeper, long enough on any conditions.

Check List

Tests

• Manual test

Related changes

• Need to cherry-pick to the release branch