Skip to content

Commit

Permalink
[#15334] yugabyted: Support for providing custom values to default po…
Browse files Browse the repository at this point in the history 
…rts in Multi-Region/Zone setups.

Summary: Creating multi-region/zone YugabyteDB clusters fail when using non-default port settings(custom port values). The issue arises because certain ports such as `master_rpc_port`, `tserver_webserver_port`, and `master_webserver_port` are hardcoded, leading to connectivity problems during cluster setup. Changes made to address this limitation by allowing these previously hardcoded ports to dynamically adjust based on the user's custom configurations either through config file or port flags.

Test Plan: Manual Testing

Reviewers: sgarg-yb

Reviewed By: sgarg-yb

Subscribers: yugabyted-dev, shikhar.sahay

Differential Revision: https://phorge.dev.yugabyte.com/D30359
  • Loading branch information
@ShikharSahay
ShikharSahay committed Nov 23, 2023
1 parent 1c46fc7 commit e88af81
Showing 1 changed file with 24 additions and 16 deletions.
40 changes: 24 additions & 16 deletions bin/yugabyted
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2717,12 +2717,12 @@ class ControlScript(object):
return jsonResponseFromMaster
except HTTPError as http_err:
Output.log("HTTP Error occured while hitting the api endpoint " +
"http://{}:7000/api/v1/tablet-servers: {}".format(master_hostport, http_err))
"http://{}/api/v1/tablet-servers: {}".format(master_hostport, http_err))
Output.log_error_and_exit(Output.make_red("ERROR:") + " Master node " +
"present at {} is not reachable.".format(master_hostport))
except Exception as err:
Output.log("HTTP Error occured while hitting the api endpoint " +
"http://{}:7000/api/v1/tablet-servers: {}".format(master_hostport, err))
"http://{}/api/v1/tablet-servers: {}".format(master_hostport, err))
Output.log_error_and_exit(Output.make_red("ERROR:") + " Master node " +
"present at {} is not reachable.".format(master_hostport))

Expand Down Expand Up @@ -2821,8 +2821,9 @@ class ControlScript(object):
# add a new master computed during configure data_placement command
def add_master_for_data_placement(self, master_addr, new_master):

new_master_rpc_port = self.configs.saved_data.get("master_rpc_port")
Output.log("Adding master {} to the cluster.".format(new_master))
if not YBAdminProxy.add_master(master_addr, new_master):
if not YBAdminProxy.add_master(master_addr, new_master, new_master_rpc_port):
Output.log_error_and_exit("Failed to add master {} to the cluster."
.format(new_master))
if not self.wait_get_master(new_master):
Expand All @@ -2834,8 +2835,9 @@ class ControlScript(object):
# remove an master computed during configure data_placement command
def remove_master_for_data_placement(self, master_addr, old_master):

old_master_rpc_port = self.configs.saved_data.get("master_rpc_port")
Output.log("Removing master {} from the cluster.".format(old_master))
if not YBAdminProxy.remove_master(master_addr, old_master):
if not YBAdminProxy.remove_master(master_addr, old_master, old_master_rpc_port):
Output.log_error_and_exit("Failed to remove master {} from the cluster."
.format(old_master))
if not self.wait_remove_master(old_master):
Expand Down Expand Up @@ -3523,11 +3525,12 @@ class ControlScript(object):
# The master was not in the current list of masters
# and we have a valid join_ip
advertise_ip = self.advertise_ip()
master_rpc_port = self.configs.saved_data.get("master_rpc_port")

if len(master_uuids) >= 3:
# this is going to be a standalone shell master
return True
if not YBAdminProxy.add_master(master_addrs, advertise_ip):
if not YBAdminProxy.add_master(master_addrs, advertise_ip, master_rpc_port):
Output.log_error_and_exit("Unable to add master {} to existing cluster at {}.".format(
advertise_ip, join_ip))
return False
Expand Down Expand Up @@ -4273,8 +4276,10 @@ class ControlScript(object):
# Get the current master leader known to a tserver using api/v1/masters endpoint
# of the tserver.
def get_current_master_leader_from_api(self, tserverIP, timeout=60):
tserver_addr = "{}:{}".format(tserverIP,
self.configs.saved_data.get("tserver_webserver_port"))
try:
tserverMastersAPI = "http://{}:9000/api/v1/masters".format(tserverIP)
tserverMastersAPI = "http://{}/api/v1/masters".format(tserver_addr)
Output.log("Trying to get masters information from {}".format(tserverMastersAPI) +
" (Timeout={})".format(timeout))
response = urlopen(Request(tserverMastersAPI), timeout=timeout)
Expand All @@ -4295,17 +4300,17 @@ class ControlScript(object):

except HTTPError as http_err:
Output.log("HTTP Error occured while hitting the api endpoint " +
"http://{}:9000/api/v1/masters: {}".format(tserverIP, http_err))
"http://{}/api/v1/masters: {}".format(tserver_addr, http_err))
Output.log_error_and_exit(Output.make_red("ERROR:") + " Node at the join ip " +
"provided is not reachable.")
except URLError as url_error:
Output.log("Some error occured while hitting the api endpoint " +
"http://{}:9000/api/v1/masters: {}".format(tserverIP, url_error))
"http://{}/api/v1/masters: {}".format(tserver_addr, url_error))
Output.log_error_and_exit(Output.make_red("ERROR:") + " Node at the join ip " +
"provided is not reachable.")
except Exception as err:
Output.log("Error occured while hitting the api endpoint " +
"http://{}:9000/api/v1/masters: {}".format(tserverIP, err))
"http://{}/api/v1/masters: {}".format(tserver_addr, err))
Output.log_error_and_exit(Output.make_red("ERROR:") + " Node at the join ip " +
"provided is not reachable.")

Expand Down Expand Up @@ -4675,7 +4680,8 @@ class ControlScript(object):
args.join = master_leader

# Get info on all tservers
master_leader_hostport = "{}:7000".format(master_leader)
master_leader_hostport = "{}:{}".format(master_leader,
self.configs.saved_data.get("master_webserver_port"))
tservers_info = dict(self.get_all_tserver_info(master_leader_hostport))

# Check if any existing node has the same IP as advertise address
Expand Down Expand Up @@ -4739,7 +4745,8 @@ class ControlScript(object):

if args.insecure:
if args.join:
master_hostport = "{}:7000".format(args.join)
master_hostport = "{}:{}".format(args.join,
self.configs.saved_data.get("master_webserver_port"))
if self.is_leader_master_secure(master_hostport):
# Case Scenario: When a User starts the 1st node in secure mode and tries
# to start the second node in insecure mode
Expand All @@ -4749,7 +4756,8 @@ class ControlScript(object):
"secure and an insecure node.")
elif args.secure:
if args.join:
master_hostport = "{}:7000".format(args.join)
master_hostport = "{}:{}".format(args.join,
self.configs.saved_data.get("master_webserver_port"))
if not self.is_leader_master_secure(master_hostport):
# Case Scenario: When the user starts the 1st node in insecure mode and
# tries to start the second node in secure mode.
Expand Down Expand Up @@ -5714,16 +5722,16 @@ class YBAdminProxy(object):
YBAdminProxy.cmd_args.append('--certs_dir_name={}'.format(certs_dir_name[0].group(1)))

@staticmethod
def add_master(master_addrs, new_master_ip, timeout=10):
def add_master(master_addrs, new_master_ip, new_master_rpc_port, timeout=10):
cmd = YBAdminProxy.cmd_args + ["--init_master_addrs", master_addrs,
"change_master_config", "ADD_SERVER", new_master_ip, "7100"]
"change_master_config", "ADD_SERVER", new_master_ip, str(new_master_rpc_port)]
out, err, ret_code = run_process(cmd, timeout=timeout, log_cmd=True)
return (0 == ret_code)

@staticmethod
def remove_master(master_addrs, old_master, timeout=10):
def remove_master(master_addrs, old_master, old_master_rpc_port, timeout=10):
cmd = YBAdminProxy.cmd_args + ["--init_master_addrs", master_addrs,
"change_master_config", "REMOVE_SERVER", old_master, "7100"]
"change_master_config", "REMOVE_SERVER", old_master, str(old_master_rpc_port)]
out, err, ret_code = run_process(cmd, timeout=timeout, log_cmd=True)
return (0 == ret_code)

Expand Down

0 comments on commit e88af81

Please sign in to comment.