Table of Contents |
---|
Linux cluster
To service multiple FIX sessions reliably, FIXEdge can be deployed in a Linux HA cluster (Red Hat High-Availability Add-on Overview) with 2 or 3 nodes and shared storage for keeping sessions a session's state.
The possible configurable options of shared state are:
- For a shared physical device
- For a shared operational system (Gluster FS)
- FIXEdge Logs Replicator
- DRBD
Cluster The cluster solution utilizes Corosync and Pacemaker – tools that facilitate the HA Linux cluster for applications that do not have native support for clustering.
The This article describes the Active-Passive / Hot-Warm Failover Cluster.
- Active or Hot node - it is a working FIXEdge node getting runtime messages
- Passive or Warm node - it is a FIXEdge node getting periodic updates from the an active one, and ready to start if the main one will failfails.
Virtual IP
FIXEdge nodes use virtual IP as an entry point for FIX Clients and remote administrating tools (for example FIXICC). Virtual IP is assigned to the node when it is active.
This approach gives a working working, highly-available solution and works well for active-passive clusters with few nodes and static FIX session configuration.
Health checks
Health checks are used to figure out if a node or an application that runs on the node is operating properly.
The simplest way to do a node health check is to monitor the FIXEdge PID file.
More precise checks:
- Check the system status via FIXEdge Admin REST API by sending the special GET Request
- Establish FIX Admin monitoring session.
Shared physical device
Shared storage might be a SAN-attached device, Fibre channel attached device or TCP/IP attached device. The device might be attached to all nodes simultaneously, or the cluster resource manager can attach it to the active node only. The device might, in turn, be in turn a resilient one, presenting the distributed file system with software or hardware replication between filesystem nodes. In the case of a geographically distributed cluster, the shared storage also can be distributed geographically in the same way as cluster nodes, and the cluster resource manager can attach the storage instance to the node the storage instance in the same geo-location.
Two or three nodes constitute a cluster and at any moment only one of them runs can be running on FIXEdge. When a failure of the active FIXEdge node is detected, the shared FIX message file system storage is unmounted on that node and mounted on the second node. Then, then FE is started on the second node. All sessions are handled by one active server and will be started on another node in case of failure.
FIXEdge start-up time grows increases with the number of the session it serves.
Info |
---|
The current approach prevents load balancing between the cluster nodes. The problem with balancing the load is resolved in this solution: FIXEdge NGE |
...
- The simplest is to use a shared network filesystem (e.g. NFS share)
The drawback is significantly increased latency (approx 20 times slower). - The recommended configuration is to use a SAN storage that can be attached to every each of the cluster nodes.
The storage itself is a block device that is mounted on the node where the FIXEdge is currently runrunning.
This approach allows having shared storage for all nodes while keeping I/O latency low.
...
- Network:
- The channel between FE and data provider - max wide
- dedicated network for cluster synchronization (heartbeats)
- Open ports:
high-availability
service- 8005/tcp
- 8901/tcp
- 8905/tcp
- 1234/udp
- Application:
- 1 Core per 1 Data Provider (recommended for latency)
- 1G for binaries
- Collocated nodes in one DC for max performance
- Health-check interval sufficient for FIXEdge to start
- Storage:
- Mandatory Fibre Channel SAN storage for session logs - 1.5 TB, directly attached to both nodes. Encrypted via standard LVM features.
- Mandatory STONITH to ensure that FC storage is mounted on a single node
- Mandatory Archive for session logs - https://kbb2bits.b2bitsatlassian.comnet/wiki/display/B2BITS/FIXEdge+Capacity
- Shared storage (NFS/SMB) for configuration - can be slow.
- Experimental features:
- Start/stop from FIXICC - concept conflicts with cluster control.
- Operating system: RHEL 7 or newer.
...
- Start FIXEdge at start-of-day - via pacemaker per FE resource by enabling resources (fixegde).
- Stop FIXEdge at end-of-day, for maintenance hours - via pacemaker by disabling resources (fixegde).
- Log archiving - mandatory for FIXEdge operation, FIXEdge needs available space on on the
/opt/fixedge/FIXEdge1/log
directory, so a periodic archiving of the files and directory cleanup is required.
...
Install packages from a repository:
Code Block language bash yum install corosync pcs pacemaker
Set the password for the
hacluster
user:Code Block language bash passwd hacluster
Open ports on the firewall:
Code Block language bash firewall-cmd --add-service=high-availability firewall-cmd --runtime-to-permanent
Enable cluster services to run at the system start-up:
Code Block language bash systemctl enable pcsd corosync pacemaker
...
To install FIXEdge you first need the following artifacts:
- the FIXEdge package (
fixedge.tar.gz
) - FIXEdge
systemd
integration configuration (fixedge-systemd.tar.gz
) - the license (
engine.license
).
Superuser privileges are required for all the steps.
Install the FIXEdge RPM package:
Code Block language bash mkdir --parents /opt/fixedge tar --extract --file fixedge.tar.gz --directory /opt/fixedge
Unpack the FIXEdge
systemd
integration configuration:Code Block language bash tar --extract --file fixedge-systemd.tar.gz --directory /
Add a user and a group for the FIXEdge:
Code Block language bash groupadd --system fixedge useradd --system --gid fixedge --home-dir /opt/fixedge --shell /sbin/nologin --comment "Account to own and run FIXEdge" fixedge
Change ownership of the FIXEdge to the dedicated user:
Code Block language bash chown --recursive fixedge:fixedge /opt/fixedge
Copy the license:
Code Block language bash cp engine.license /opt/fixedge
Enable the FIXICC Agent to start at the system start-up:
Code Block systemctl enable fixicc-agent
...
Prepare Storage for the Session Logs and the Configuration
At this point, FIXEdge is deployed locally. Now, now we need to make the configuration and the state shared.
...
Create a filesystem on the logical volume:
Code Block language bash mkfs -t xfs /dev/shared_logs_group/shared_logs
Mount the filesystem:
Code Block language bash mount /dev/shared_logs_group/shared_logs /opt/fixedge/FIXEdge1/log
Create a FIXEdge log directory structure:
Code Block language bash mkdir /opt/fixedge/FIXEdge1/log/archive /opt/fixedge/FIXEdge1/log/backup
Unmount and deactivate:
Code Block language bash umount /opt/fixedge/FIXEdge1/log vgchange -a n shared_logs_group
Set up a FIXEdge Cluster
Now we need to set up cluster resources for the FIXEdge. Superuser privileges are required for all the steps.
- Authorize nodes for
hacluster
user/log vgchange -a n shared_logs_group
Set up a FIXEdge Cluster
Now we need to set up cluster resources for FIXEdge. Superuser privileges are required for all the steps.
Authorize nodes for
hacluster
user:Code Block language bash pcs cluster auth NODE_1_NAME NODE_2_NAME -u hacluster
Where
NODE_1_NAME
andNODE_2_NAME
are the hostnames of the servers that run FIXEdge.Info Note that these names must not resolve to 127.0.0.1
locally or use IP addresses instead of the hostnames.Create the cluster and add nodes:
Code Block language bash pcs cluster auth NODE setup --force --name fixedge_ha NODE_1_NAME NODE_2_NAME -u hacluster NAME
Where
NODE_1_NAME
and
areNODE_2_NAME
are the hostnames of the servers
whichthat run FIXEdge.
Create the cluster and add nodesInfo Note that these names must not resolve to 127.0.0.1
locally or use IP addresses instead of the hostnames.of the hostnames. Start cluster
Code Block language bash pcs cluster start --all
Now the cluster is starting up, you can check its status with commands:
Code Block language bash pcs status cluster pcs status nodes
Disable resource migration on the first failure, since restarting on the same node takes less time than migration to another node:
Code Block language bash pcs clusterproperty setupset start-failure-force --name fixedge_ha NODE_1_NAME NODE_2_NAME
Where
NODE_1_NAME
andNODE_2_NAME
are the hostnames of the servers which run FIXEdge.Info Note that these names must not resolve to 127.0.0.1
locally or use IP addresses instead of the hostnames.Start cluster
is-fatal=false
For a two-node cluster we must disable the quorum, but do not do this for a three-node cluster. Quorum avoids the situation when the cluster cannot decide which node is active.
Now the cluster is starting up, you can check its status with commandsCode Block language bash pcs clusterproperty startset no-quorum-all
policy=ignore
Add a virtual IP as a resource to the cluster:
Disable resource migration on the first failure, since restarting on the same node takes less time than migration to another nodeCode Block language bash pcs status cluster pcs status nodes
resource create virtual_ip ocf:heartbeat:IPaddr2 ip=<VIRTUAL_IP> cidr_netmask=32 op monitor interval=30s
where
<VIRTUAL_IP>
is the IP that will be used by the FIX clients to connect to the cluster.
Use a Dual-Port SAN Device
Add an LVM group for the sessions' logs as a resource to the cluster:
For a two-nodes cluster we must disable the quorum, do not do this for a three-node cluster. Quorum avoids the situation when the cluster cannot decide which node is active.Code Block language bash pcs property set start-failure-is-fatal=false
resource create logs_vg ocf:heartbeat:LVM volgrpname=logs
Add a filesystem for the sessions' logs as a resource to the cluster:
Add a virtual IPCode Block language bash pcs property set no-quorum-policy=ignore
resource create logs_fs ocf:heartbeat:Filesystem device=/dev/shared_logs_group/shared_logs directory=/opt/fixedge/FIXEdge1/log fstype=xfs
Add FIXEdge as a resource to the cluster:
Code Block language bash pcs resource create virtual_ip ocf:heartbeat:IPaddr2 ip=<VIRTUAL_IP> cidr_netmask=32fixedge systemd:fixedge op start timeout=300s op stop timeout=60s op monitor interval=30s
where
<VIRTUAL_IP>
is the IP that will be used by the FIX clients to connect to the cluster.
Use a Dual-Port SAN Device
- Add LVM group for the sessions' logs as a resource to the cluster
10 timeout=60s meta migration-threshold=3
Make sure that all resources are started on the same node:
Code Block language bash pcs resource create logs_vg ocf:heartbeat:LVM volgrpname=logs
Add filesystem for the sessions' logs as a resource to the cluster:
constraint colocation set virtual_ip logs_vg logs_fs fixedge sequential=true setoptions score=INFINITY
Make sure that resources are started in the proper order:
Code Block language bash pcs constraint order set virtual_ip logs_vg logs_fs fixedge
Starting and stopping FIXEdge cluster resource from FIXICC
To avoid issues related to unexpected behavior by cluster software the user should prepare scripts that use cluster management commands.
Create scripts:
bin/FixEdge1.run.cluster.sh
Add FIXEdge as a resource to the cluster:Code Block language bash title bin/FixEdge1.run.cluster.sh pcs resource create logs_fs ocf:heartbeat:Filesystem device=/dev/shared_logs_group/shared_logs directory=/opt/fixedge/FIXEdge1/log fstype=xfs
enable fixedge
This script is used for starting FIXEdge service as a cluster resource
bin/FixEdge1.stop.cluster.sh
Code Block language bash title bin/FixEdge1.stop.cluster.sh pcs resource create fixedge systemd:fixedge op start timeout=300s op stop timeout=60s op monitor interval=10 timeout=60s meta migration-threshold=3
Make sure that all resources are started on the same node:
Code Block language bash pcs constraint colocation set virtual_ip logs_vg logs_fs fixedge sequential=true setoptions score=INFINITY
Make sure that resources are started in the proper order:
Code Block language bash pcs constraint order set virtual_ip logs_vg logs_fs fixedge
disable fixedge
This script is used for stopping FIXEdge service as a cluster resource
Update the paths to start and stop scripts in the fixicc-agent/conf/agent.properties.
Code Block | ||
---|---|---|
| ||
StartFile = bin/FixEdge1.run.sh
StopFile = bin/FixEdge1.stop.sh |
The user who runs fixicc-agent should have permission to operate the cluster. In some cases, it is required to add sudo for running pcs commands
Info |
---|
The script will log to console so usually for troubleshooting and debugging purposes the scripts should be extended with the logic of forwarding standard and error output with timestamps to some file |
How to Validate the Installation
Use FIXICC to connect to the cluster nodes: add servers with addresses of the cluster nodes. The FIXICC will show one server running and another stopped.
...
Open https://NODE_1_NAME:2224/ to access the pacemaker's WebUI. It provides means for inspecting the state of the cluster nodes and resources. Under normal operation, all nodes and resources must be running. All resources must be running the same node.
...
In this setup, the TCP/IP attached device is used. A two nodes -node GlusterFS file server with mirroring is used as the shared device. The device is mounted to both nodes simultaneously. See the diagram below:
...
Do these steps on both servers: NODE_1_NAME and NODE_2_NAME
Download and install:
Code Block language bash $ sudo wget -P /etc/yum.repos.d http://download.gluster.org/pub/gluster/glusterfs/LATEST/CentOS/glusterfs-epel.repo $ sudo yum install glusterfs $ sudo yum install glusterfs-fuse $ sudo yum install glusterfs-server
Check installed version:
Code Block language bash $ glusterfsd --version glusterfs 3.6.2 built on Jan 22 2015 12:58:10 Repository revision: git://git.gluster.com/glusterfs.git Copyright (c) 2006-2013 Red Hat, Inc. <http://www.redhat.com/> GlusterFS comes with ABSOLUTELY NO WARRANTY. It is licensed to you under your choice of the GNU Lesser General Public License, version 3 or any later version (LGPLv3 or later), or the GNU General Public License, version 2 (GPLv2), in all cases as published by the Free Software Foundation.
Start glusterfs services on all servers with and enable them to start automatically on startup:
Code Block language bash $ sudo /etc/init.d/glusterd start $ sudo chkconfig glusterfsd on
...
Setting up FIXEdge instances
Copy FixEdgeCopy the FixEdge-*.tar.gz Linux package to /home/user to both nodes: NODE_1_NAME and NODE_2_NAME
...
Unpack the FIXEdge-x.x.x.tar.gz archive to /home/user/
Code Block language bash $ tar xf FixEdge-5.8.2.68334-Linux-2.6.32-gcc447-x86_64.tar -C /home/user/
The installation directory will be /home/user/FIXEdge:
- Copy the license 'engine.license to license' to the /home/user/FIXEdge folder.
Move FIXEdge configuration and logs directory to shared storage mounted to /data:
Code Block language bash mv FIXEdge1 /data/
- Edit scripts in /home/user/FIXEdge/bin to direct them to the new FIXEdge instance location:
- replace ".." with "/data".
- Edit FIXICC Agent configuration:
- /home/user/FIXEdge/fixicc-agent/conf/wrapper.conf
- wrapper.daemon.pid.dir = ${wrapper_home}
- /home/user/FixEdge/fixicc-agent/conf/agent.properties
- EngineProperty = /data/FIXEdge1/conf/engine.properties
- FIXEdgeFileSettings = /data/FIXEdge1/conf/FIXEdge.properties
- LogUrl = /data/FIXEdge1/log
- /home/user/FIXEdge/fixicc-agent/conf/wrapper.conf
- Edit FIXEdge and engine configuration:
- /data/FixEdge1/conf/engine.properties
- EngineRoot = /data
- LicenseFile = /home/user/FIXEdge/engine.license
- /data/FixEdge1/conf/FIXEdge.properties
- FIXEdge.RootDir = /data/FIXEdge1
- Log.File.RootDir = /data/FIXEdge1
- TransportLayer.SmtpTA.DllName = /home/user/FIXEdge/bin/libSMTPTA.so
- /data/FixEdge1/conf/engine.properties
Install and Start FIXICC Agent daemon:
Code Block language bash $ cd /home/user/FIXEdge/fixicc-agent/bin $ ./installDaemon.sh $ ./startDaemon.sh
- Now everything is ready to run FIXEdge on Node 1.
Prepare to copy the installation to Node 2:
Code Block language bash $ cd /home/user $ tar cvf FixEdge.tar FixEdge $ gzip FixEdge.tar
- Copy file FixEdge.tar.gz to Node2:/user/home
...
Unpack the FIXEdge-x.x.x.tar.gz archive to /home/user/
Code Block language bash $ tar xf FixEdge-5.8.2.68334-Linux-2.6.32-gcc447-x86_64.tar -C /home/user/
Install and Start the FIXICC Agent daemon:
Code Block language bash $ cd /home/user/FIXEdge/fixicc-agent/bin $ ./installDaemon.sh $ ./startDaemon.sh
...
https://github.com/feist/pcs/issues/
On both nodes, install the needed software:
Code Block language bash $ sudo yum install corosync pcs pacemaker
On both nodes, set the password for hacluster user ('epmc-cmcc' was used):
Code Block language bash $ sudo passwd hacluster
Configure Firewall on both nodes to allow cluster traffic:
Code Block language bash $ sudo iptables -I INPUT -m state --state NEW -p udp -m multiport --dports 5404,5405 -j ACCEPT $ sudo iptables -I INPUT -p tcp -m state --state NEW -m tcp --dport 2224 -j ACCEPT $ sudo iptables -I INPUT -p igmp -j ACCEPT $ sudo iptables -I INPUT -m addrtype --dst-type MULTICAST -j ACCEPT $ sudo service iptables save
Start the pcsd service on both nodes:
Code Block language bash $ sudo systemctl start pcsd
From now on, all commands needs need to be executed on one node only. We can control the cluster by using PCS from on one of the nodes.
Since we will configure all nodes from one point, we need to authenticate on all nodes before we are allowed to change the configuration. Use the previously configured hacluster user and password to do this:Code Block language bash $ sudo pcs cluster auth NODE_1_NAME NODE_2_NAME Username: hacluster Password: NODE_1_NAME: Authorized NODE_2_NAME: Authorized
Create the cluster and add nodes. This command creates the cluster node configuration in /etc/corosync.conf.
Code Block language bash $ sudo pcs cluster setup --name fixedge_cluster NODE_1_NAME NODE_2_NAME Shutting down pacemaker/corosync services... Redirecting to /bin/systemctl stop pacemaker.service Redirecting to /bin/systemctl stop corosync.service Killing any remaining services... Removing all cluster configuration files... NODE_1_NAME: Succeeded NODE_2_NAME: Succeeded
We can start cluster now:
Code Block language bash $ sudo pcs cluster start --all NODE_1_NAME: Starting Cluster... NODE_2_NAME: Starting Cluster...
We can check cluster status:
Code Block language bash $ sudo pcs status cluster Cluster Status: Last updated: Tue Jan 27 22:11:15 2015 Last change: Tue Jan 27 22:10:48 2015 via crmd on NODE_1_NAME Stack: corosync Current DC: NODE_1_NAME (1) - partition with quorum Version: 1.1.10-32.el7_0.1-368c726 2 Nodes configured 0 Resources configured $ sudo pcs status nodes Pacemaker Nodes: Online: NODE_1_NAME NODE_2_NAME Standby: Offline: $ sudo corosync-cmapctl | grep members runtime.totem.pg.mrp.srp.members.1.config_version (u64) = 0 runtime.totem.pg.mrp.srp.members.1.ip (str) = r(0) ip(10.17.131.127) runtime.totem.pg.mrp.srp.members.1.join_count (u32) = 1 runtime.totem.pg.mrp.srp.members.1.status (str) = joined runtime.totem.pg.mrp.srp.members.2.config_version (u64) = 0 runtime.totem.pg.mrp.srp.members.2.ip (str) = r(0) ip(10.17.131.128) runtime.totem.pg.mrp.srp.members.2.join_count (u32) = 1 runtime.totem.pg.mrp.srp.members.2.status (str) = joined $ sudo pcs status corosync Membership information ---------------------- Nodeid Votes Name 1 1 NODE_1_NAME (local) 2 1 NODE_2_NAME
Disable the STONITH option as we don't have STONITH devices in our demo virtual environment:
Code Block language bash $ sudo pcs property set stonith-enabled=false
For a two-nodes node cluster we must disable the quorum:
Code Block language bash $ sudo pcs property set no-quorum-policy=ignore $ sudo pcs property Cluster Properties: cluster-infrastructure: corosync dc-version: 1.1.10-32.el7_0.1-368c726 no-quorum-policy: ignore stonith-enabled: false
Add Virtual IP as a resource to the cluster:
Code Block language bash $ sudo pcs resource create virtual_ip ocf:heartbeat:IPaddr2 ip=10.17.135.17 cidr_netmask=32 op monitor interval=30s $ sudo pcs status resources virtual_ip (ocf::heartbeat:IPaddr2): Started
Add FIXEdge as a resource to cluster:
Code Block language bash $ sudo pcs resource create FIXEdge ocf:heartbeat:anything params binfile="/home/user/FixEdge/bin/FIXEdge" cmdline_options="/data/FixEdge1/conf/FIXEdge.properties" user="user" logfile="/home/user/FIXEdge_resource.log" errlogfile="/home/user/FIXEdge_resource_error.log"
Note For some reason in the /usr/lib/ocf/resource.d/ of the installed cluster there are many missing agents, including ocf:heartbeat:anything. You need to modify the original version (which you can download here: https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/anything) to make it working. The working version of the agent is attached.
This file should be copied to /usr/lib/ocf/resource.d/ and make executable:
Code Block language bash $ sudo cp anything /usr/lib/ocf/resource.d/heartbeat/ $ sudo chmod a+rwx /usr/lib/ocf/resource.d/heartbeat/anything
Also, to make this agent works the following lines shall be added to sudoers file:
Code Block language bash $ sudo visudo Defaults !requiretty user ALL=(user) NOPASSWD: ALL root ALL=(user) NOPASSWD: ALL
In order to make sure that the Virtual IP and FIXEdge always stay together, we can add a constraint:
Code Block language bash $ sudo pcs constraint colocation add FIXEdge virtual_ip INFINITY
To avoid the situation where the FIXEdge would start before the virtual IP is started or owned by a certain node, we need to add another constraint which that determines the order of availability of both resources:
Code Block language bash $ sudo pcs constraint order virtual_ip then FIXEdge Adding virtual_ip FIXEdge (kind: Mandatory) (Options: first-action=start then-action=start)
After configuring the cluster with the correct constraints, restart it and check the status:
Code Block language bash $ sudo pcs cluster stop --all && sudo pcs cluster start --all NODE_1_NAME: Stopping Cluster... NODE_2_NAME: Stopping Cluster... NODE_2_NAME: Starting Cluster... NODE_1_NAME: Starting Cluster...
- The cluster configuration is now completed.
...
File | Size | Created | Comment |
---|---|---|---|
587268 | Mar 06, 2015 21:49 | Modified Sender source code (Sending orders with incremented ClOrdID) | |
9475 | Jan 28, 2015 09:16 | ocf:heartbeat:anything agent for FIXEdge resource monitoring | |
4976201 | Mar 08, 2015 01:06 | Clients simulator (binary) | |
34 KB | Jan 20, 2020 |
FIX Logs Replicator
The FIX Antenna Replication tool (RT) is typically used for FIX session real-time replication on a backup host with FIX engine.
...
The current article describes the Recovery Time Objective (RTO) and Recovery Point Objective(RPO) for Disaster recovery in case of active-passive Cluster configuration (see FIXEdge Failover Cluster installation).
Recovery Time Objective (RTO)
...
Failure of software or hardware within a node (e.g. application failure, node unavailability, resource exhausting). Failover Cluster consists of several application nodes. In case of one node fails – , the application (or DB) is moved to another node and starts there within approximately 2 minutes.
The session recovery procedure happens automatically. The missing messages should be recovered with a resend request procedure automatically.
Info |
---|
Session recovery requires a reset sequence in case of damaged storage. The messages from the beginning of the day will be lost when resetting the sequence. See Recovery procedure for a session with corrupted storagesIt is possible to manually request missing messages since sequence number equal one using the Logon (A) with the desired sequence and the subsequent Request Request (2) messages (see the FIX specification https://www.fixtrading.org/standards/) FIX Standard recommends requesting sequences after logon using Message recovery procedure or use Extended features for FIX session and FIX connection initiation |
Achievable RTO Values: "Greater than 0 seconds, up to and including 2 minutes"
...
Achievable RPO Values: "Recent synch point: Greater than 0 seconds, up to and including 1 day, i.e. since the start of the day".
Info |
---|
The better Better RPO values can be achieved with additional configuration or tools like the log replication tool. This procedure requires additional configuration for each client individually. |
...