2 region 3 DC config issue

(binzhang) #1

I’m trying to config a 3 DC, 2 region cluster.

  • Region A ( write ) = DC1 and DC2
  • Region B = DC3

After i do some data load test, i can find data file / log file is large on DC1 and DC2. But DC3 's data file and log file size is very very small. so i suspect data is not replicated to region B.

Below is my config. can you please suggest where 's wrong?

VM                  datacenter_id      $ID  class    $ID  class    $ID  class     $ID   class
----------------------------- --------- ------ --------- ----------- --------------- --------------------
pgc3lvsprojecta01-555694.stratus.   dc1    4500 stateless 4501 storage  4502 storage   4505 transaction 
pgc3lvsprojecta02-555705.stratus.   dc1    4500 stateless 4501 storage  4502 storage   4505 transaction 
pgc3lvsprojecta03-555706.stratus.   dc1    4500 stateless 4503 transaction  4505 stateless 
pgc3lvsprojecta04-555707.stratus.   dc2    4500 stateless 4503 transaction  4505 stateless 
c3lvspgdbbztest00-357425.stratus.   dc2    4500 stateless 4503 transaction  4505 stateless 
c3lvspgdbbztest01-357362.stratus.    dc3    4500 stateless 4501 storage  4502 storage 
c3lvspgdbbztest02-357373.stratus.    dc3    4500 stateless 4501 storage  4502 storage 
c3lvspgdbbztest03-357424.stratus.    dc3    4500 stateless 4503 transaction  4505 stateless 

Sample conf file on one machine, other machine’s difference is at datacenter_id and [fdbserver.$ID]

[root@pgc3lvsprojecta01-555694 foundationdb]# cat foundationdb.conf

[fdbmonitor]
user = foundationdb
group = foundationdb

[general]
restart_delay = 60
## by default, restart_backoff = restart_delay_reset_interval = restart_delay
# initial_restart_delay = 0
# restart_backoff = 60
# restart_delay_reset_interval = 60
cluster_file = /etc/foundationdb/fdb.cluster
# delete_envvars =
# kill_on_configuration_change = true

## Default parameters for individual fdbserver processes
[fdbserver]
command = /usr/sbin/fdbserver
public_address = auto:$ID
listen_address = public
datadir = /var/lib/foundationdb/data/$ID
logdir = /var/log/foundationdb
# logsize = 10MiB
# maxlogssize = 100MiB
# machine_id =
datacenter_id = dc1
# class =
# memory = 8GiB
# storage_memory = 1GiB
# metrics_cluster =
# metrics_prefix =

## An individual fdbserver process with id 4500
## Parameters set here override defaults from the [fdbserver] section
[fdbserver.4500]
class = stateless
[fdbserver.4501]
class = storage
[fdbserver.4502]
class = storage
[fdbserver.4505]
class = transaction
[backup_agent]
command = /usr/lib/foundationdb/backup_agent/backup_agent
logdir = /var/log/foundationdb

[backup_agent.1]
fdb> status

Using cluster file `fdb.cluster'.

Configuration:
  Redundancy mode        - double
  Storage engine         - ssd-2
  Coordinators           - 5
  Desired Proxies        - 3

Cluster:
  FoundationDB processes - 28
  Machines               - 8
  Memory availability    - 5.4 GB per process on machine with least available
  Fault Tolerance        - 1 machine
  Server time            - 05/03/19 15:03:40

Data:
  Replication health     - Healthy (Rebalancing)
  Moving data            - 0.000 GB
  Sum of key-value sizes - 818 MB
  Disk space used        - 2.383 GB
  "cluster" : {
        "clients" : {
            "count" : 0
        },
        "cluster_controller_timestamp" : 1556921091,
        "configuration" : {
            "coordinators_count" : 5,
            "excluded_servers" : [
            ],
            "proxies" : 3,
            "redundancy_mode" : "double",
            "regions" : [
                {
                    "datacenters" : [
                        {
                            "id" : "dc1",
                            "priority" : 1
                        },
                        {
                            "id" : "dc2",
                            "priority" : 0,
                            "satellite" : 1
                        }
                    ],
                    "satellite_redundancy_mode" : "one_satellite_double"
                },
                {
                    "datacenters" : [
                        {
                            "id" : "dc3",
                            "priority" : 0
                        }
                    ]
                }
            ],
            "storage_engine" : "ssd-2",
            "usable_regions" : 1
        },

machine c3lvspgdbbztest03-357424 belong to dc3, but it storage server process’s data dir don’t have any data file.

[root@c3lvspgdbbztest03-357424 foundationdb]# ps -ef |grep class
foundat+ 15776 15774  1 10:30 ?        00:03:30 /usr/sbin/fdbserver --class stateless --cluster_file /etc/foundationdb/fdb.cluster --datacenter_id dc3 --datadir /var/lib/foundationdb/data/4500 --listen_address public --logdir /var/log/foundationdb --public_address auto:4500
foundat+ 15777 15774  0 10:30 ?        00:02:06 /usr/sbin/fdbserver --class transaction --cluster_file /etc/foundationdb/fdb.cluster --datacenter_id dc3 --datadir /var/lib/foundationdb/data/4503 --listen_address public --logdir /var/log/foundationdb --public_address auto:4503
foundat+ 15778 15774  1 10:30 ?        00:03:08 /usr/sbin/fdbserver --class stateless --cluster_file /etc/foundationdb/fdb.cluster --datacenter_id dc3 --datadir /var/lib/foundationdb/data/4505 --listen_address public --logdir /var/log/foundationdb --public_address auto:4505
root     19772 15256  0 15:06 pts/1    00:00:00 grep --color=auto class
 
[root@c3lvspgdbbztest03-357424 foundationdb]# ls -l /var/lib/foundationdb/data/4503
total 8
-rw-r--r-- 1 foundationdb foundationdb   11 May  3 11:22 fitness
-rw---S--- 1 foundationdb foundationdb 4096 May  3 10:30 processId
[root@c3lvspgdbbztest03-357424 foundationdb]#
(Alex Miller) #2

Changing the usable_regions configuration:

The usable_regions configuration option determines the number of regions which have a replica of the database.
Increasing the usable_regions will start copying data from the active region to the remote region. Reducing the usable_regions will immediately drop the replicas in the remote region. During these changes, only one primary datacenter can have priority >= 0. This enforces exactly which region will lose its replica.

So you probably just need to configure usable_regions:=2

(binzhang) #3
fdb> configure usable_regions:=2
ERROR: When changing usable_regions, only one region can have priority >= 0
Type `configure FORCE <TOKEN>*' to configure without this check
fdb>

this is our region config json:

 {
"regions":[{
    "datacenters":[{
        "id":"dc1",
        "priority":1
    },{
        "id":"dc2",
        "priority":0,
        "satellite":1
    }],
    "satellite_redundancy_mode":"one_satellite_double"
    },
    {
    "datacenters":[{
        "id":"dc3",
        "priority":0
       }]
    }]
}
(Alex Miller) #4

Ah, I was letting you be risky, but fdbcli won’t by default.

It wants you to set dc3 to have a priority of -1, and then configure usable_regions to 2, so that the system won’t think that if dc1 fails, it could recover to dc3.

Similarly, if you wanted to go to a one region configuration, you’d first need to change the priorities to make it explicit which region you want the primary to be, and then configure usable_regions to 1.

1 Like
(binzhang) #5

i changed piority=-1 for dc3, and is able to set usable_regions=2; then i can see that data is auto streamed to region 2(dc3).

We have 3 DC, set-up 2 region. Region A include dc1 & dc2, region B include dc3. our purpose is, if either dc1 or dc2 power down, region B(dc3) become write dc/region. When dc1/dc2 power on, cluster auto swichover write node to region A(dc1 & dc2). i will do more test on this layout.

appreciate your promptly help.

(Alex Miller) #6

Once your data is replicated to the secondary, you’ll want to switch the priority back to a non-negative number, otherwise the cluster won’t switch over when the primary dies.

There’s a DatacenterVersionDifference trace event that you can use to see the version difference between the primary and secondary. If it’s under 5,000,000, then you’re probably caught up.