jrinetd

jrinetd is an open source (Apache License, Version 2.0) Java TCP port redirector proxy. Do not require any external lib.

Current Stable Version is 1.1.2

DOC

Schema about Forward / Port Redirector:

1. Machine-A (Client) init connection to Machine-B (jrinetd)
2. Machine-B init connection to Machine-C (Server)
3. Done: Machine-A is able to speak with Machine-C

Notes about security:

• Machine-A (Client) may be in Internal network.
• Machine-B (jrinetd) may be in DMZ.
• Machine-C (Server) may be in External network.

System Properties (optional)

# To redir stdout/stderr to (auto-daily-rotated) files you can use:
-Dlog.stdOutFile=/var/log/jrinetd.out -Dlog.stdErrFile=/var/log/jrinetd.err
# To log to stdout too:
-Dlog.stdToo=true 

Filenames are a base-pattern, output files they will be: jrinetd.xxx.YEAR-MONTH-DAY (jrinetd.xxx.2015-08-01)

Config (jrinetd.conf)

Config file must be in class-path ${JRINETD_HOME}/conf/, general format is:

#### Forward / Port Redirector
## forward <listen-addr>:<listen-port> <endpoint-list> [opts]

# Note: <endpoint-list> can be a coma separated list of addresses, like "srv1:80,srv2:80,10.0.0.3:8080"

Options are comma separated:

• Options for outgoing connections
  • Loadbalancing (only one option can be used)
    • LB=NONE: disable LoadBalancing
    • LB=ORDER: active LoadBalancing in order (DNS resolved IP address are sorted, lower first { 10.0.0.1, 10.0.0.2, 192.168.0.1 })
    • LB=RR: active LoadBalancing in round-robin (DNS order)
    • LB=RAND: activate LoadBalancing in random order
    • LB=RANDRR: activate LoadBalancing in random order and round-robin
  • Failover (default disabled)
    • FAILOVER: enable FailOver (if connect fail, try next address)
  • Sticky Session
    • STICKY=MEM:bitmask:elements:ttl:sticky-name[:cluster-name]: activate Sticky session based on IP Source Address. Sessions are stored in MEMory, bitmask is a CIDR to apply in source-ip-address (16=Class B, 24=Class C, 32=Unique host), elements for LRU cache, ttl is time to live of elements in cache (seconds), sticky-name and cluster-name in cluster environment is cluster identifier and replication identifier respectively.
• Options for inbound connections
  • PROXY=SEND: use PROXY protocol (v1), generate header for remote server

Example config of Forward / Port Redirector:

# <listen-addr>:<listen-port> <endpoint-list> [opts]
forward 0.0.0.0:80 10.0.0.1:8080,10.0.0.2:8080
forward 127.0.0.1:443 www.acme.com:443 LB=RR,STICKY=MEM:24:128:300:sticky1

• More examples in sampleconf

Running (Linux)

./bin/jrinetd.sh <start|stop|restart|reload|status>

TODOs

• Use Log4J
• Limit number of connections
• Limit absolute timeout/TTL of a connection
• Configurable retry-sleeps
• Thread pool/control
• Custom timeout by binding
• Audit threads / connections
• Statistics/Accounting
• JMX

DONEs

• NIO (v1.0.0)
• BufferPool for reduce GC pressure (v1.0.0)
• Reload config (v1.0.0)
• Allow alternative config names (v1.0.0)
• Zip Packaging (Maven Assembly) (v1.0.0)
• Allow redir stdout/stderr to File, with auto daily-rotate (v1.0.0)
• PROXY protocol (v1) for Outgoing connections (v1.0.0)
• Multiple endpoint-list (not only DNS multi A-record) (v1.0.0)
• Sticky sessions in LoadBalancing (v1.0.0)
• Use multiple thread for multi-core machines (v1.0.0)
• Replicate Sticky Sessions over multiple jrinetd (HA) (v1.0.0)
• Improved support for embed -a little- (v1.1.0)

MISC

Current harcoded values:

• Buffer Pool size: 8192buffers (per thread)
• Buffer-Length for I/O: 2048bytes
• DNS cache: 3seconds
• Reload config check time interval: 10seconds
• Shutdown/Reload timeout: 30seconds
• Cluster Connection timeout: 10seconds

Latency Benchmark

Throughput Benchmark

All test run on localhost on a Laptop. Values are not accurate, but orientative. Latency { EchoServer, 1 byte write/read (end-to-end, round-trip), 100K iterations } Lower Better. Throughput { Chargen, 1024bytes read & write (full-duplex), total 512MBytes } Higher better.

Inspired in rinetd, this is a Java-minimalistic version.