FLOM can operate in 4 different ways:
- local only: the lock manager operates inside a single running system, this is not a distributed feature at all
- network: the lock manager operates inside an IP network that may contain any number of running systems. At least one node must be chosen to host flom daemon, other nodes must point to a flom daemon node. This option gives you a distributed lock manager you can use it to synchronize activities happening inside different systems
- network with autodiscovery: this mode extends network mode adding the support for flom daemon autodiscovery feature. This option adds you a lot of flexibility: you can change flom daemon host without reconfiguring other systems: flom daemon is discovered using UDP/IP multicast protocol
- dynamic network: this mode gives you a zero conf distributed lock manager because the flom daemon will be automatically activated when necessary and automatically discovered by other running systems.
Dynamic network mode warning
It might seem that dynamic network is a sort of catch all solution: you could discover how easy and powerful it is just reading the related Use Case 9. Unfortunately it’s an unsafe operation mode: you should use it only if an error in synchronization does not corrupt your data and does not allows your computation to produce inconsistent results.
Repeat: dynamic network mode can corrupt your data and/or break your algorithms.
But if you are using the flom distributed lock manager to synchronize processes that can coexist without any side effect (a typical scenario is workload balancing issues), dynamic network mode can give you unbeatable flexibility and ease of use.
Why dynamic network mode is dangerous
Referring to CAP Theorem the biggest issue correlated to dynamic network mode is related to network partitions: if, for any reason, your network splits from an UDP/IP multicast point of view (some systems are not able to communicate with others using UDP/IP multicast or UDP/IP multicast are slowed down), you will have extra running flom daemon instances and your processes will synchronize in different unpredictable clusters. The depicted scenario is inconvenient and now you do know it!
FLOM does not try to implement some complex algorithm like totally ordered multicast to manage this type of issues: it simply perform a request/reply with timeout: if reply message don’t arrive before timeout expiration, the requester will activate a new flom daemon.
Local and network configuration modes
There are basically 3 different options that influence operation mode. This first table resume the command line option and configuration file keywords necessary to specify local and network modes:
| Config type | Unicast (TCP/IP) address | Multicast (UDP/IP) address | Lifespan |
|---|---|---|---|
| command line option (short form) | -a | -A | -d |
| command line option (long form) | --unicast-address | --multicast-address | --daemon-lifespan |
| config file keyword | UnicastAddress | MulticastAddress | Lifespan |
This second table explains the behavior obtained changing configuration options:
| Unicast address | Multicast address | Lifespan | Behavior |
|---|---|---|---|
| null | null | < 0 | local mode (no network): if a flom daemon is not locally available a new one will be started and it will not spontaneously end |
| null | null | = 0 | local mode (no network): if a flom daemon is not locally available a new one will not be started and the command will fail |
| null | null | > 0 | local mode (no network): if a flom daemon is not locally available a new one will be started and it will end after the inactivity time exceeds the Lifespan specified value |
| IP address/name | null | < 0 | network mode: if a flom daemon is not available at the specified IP address/name a new one will be started locally trying to bind it to the specified IP address/name; if started, the daemon would not spontaneously end. Note: IP address/name must be a valid address/name for at least one of the network interfaces locally available |
| IP address/name | null | = 0 | network mode: if a flom daemon is not available at the specified IP address/name a new one will not be started locally and the command will fail |
| IP address/name | null | > 0 | network mode: if a flom daemon is not available at the specified IP address/name a new one will be started locally trying to bind it to the specified IP address/name; if started, the daemon would spontaneously end after the inactivity time exceeds the Lifespan specified value. Note: IP address/name must be a valid address/name for at least one of the network interfaces locally available |
| IP address/name | IP address/name | < 0 | network with autodiscovery mode: if a flom daemon is not available at the specified IP address/name a new one will be started locally trying to bind it to the specified IP address/name; if started, the daemon would not spontaneously end. If activated, the daemon would answer to UDP/IP multicast query: autodiscovery feature. Note: IP address/name must be a valid address/name for at least one of the network interfaces locally available |
| IP address/name | IP address/name | = 0 | network with autodiscovery mode: if a flom daemon is not available at the specified IP address/name a new one will not be started locally and the command will fail. Note: multicast UDP/IP address will be ignored because unicast TCP/IP address is specified |
| IP address/name | IP address/name | > 0 | network with autodiscovery mode: if a flom daemon is not available at the specified IP address/name a new one will be started locally trying to bind it to the specified IP address/name; if started, the daemon would spontaneously end after the inactivity time exceeds the Lifespan specified value. If activated, the daemon would answer to UDP/IP multicast query: autodiscovery feature. Note: IP address/name must be a valid address/name for at least one of the network interfaces locally available |
| null | IP address/name | < 0 | dynamic network mode: if a flom daemon does not reply to UDP/IP multicast query at the specified IP address/name a new one will be started locally trying to bind all the available interfaces (INADDR_ANY); if started, the daemon would not spontaneously end. If activated, the daemon would answer to UDP/IP multicast query: autodiscovery feature. |
| null | IP address/name | = 0 | dynamic network mode: if a flom daemon does not reply to UDP/IP multicast query at the specified IP address/name a new one will not be started locally and the command will fail |
| null | IP address/name | > 0 | dynamic network mode: if a flom daemon does not reply to UDP/IP multicast query at the specified IP address/name a new one will be started locally trying to bind all the available interfaces (INADDR_ANY); if started, the daemon would spontaneously end after the inactivity time exceeds the Lifespan specified value. If activated, the daemon would answer to UDP/IP multicast query: autodiscovery feature |
Additional details
When operating in local mode, configuration option “-s”, “--socket-name”, “SocketName” can be used to change the default behavior; default behavior uses a per user socket name and, as a consequence, a per user flom daemon. See Use Case 6 for additional information.
The following table resumes the configuration options necessary to customize the ports used for TCP/IP unicast and UDP/IP multicast communications:
| Config type | Unicast (TCP/IP) port | Multicast (UDP/IP) port |
|---|---|---|
| command line option (short form) | -p | -P |
| command line option (long form) | --unicast-port | --multicast-port |
| config file keyword | UnicastPort | MulticastPort |
Note related to ports
Configuration options related to ports are meaningful only if the correspondent configuration options related to addresses are set.
flom man page is the official location to discover the default port for your flom version.
UDP/IP multicast and firewalls
Please pay attention some Linux distros configure a default firewall that silently drop multicast datagrams: disable your firewall or configure it to not drop multicast datagrams. openSUSE 12.2 is such a distribution.