Diagnosing Java Runtime Problems: CPU, Memory, Disk, and Network Tips

Network Troubleshooting

Distinguish connection timeout ( connectionTimeout) from read/write timeout ( readTimeout / writeTimeout). Keep client‑side timeouts shorter than server limits.

TCP Queue Overflow

Monitor SYN and accept queues:

netstat -s | egrep "listen|LISTEN"

ss -lnt

“overflowed” counts full accept queues; “sockets dropped” counts SYN‑queue overflow.

RST Packets

RST indicates abrupt termination. Common causes:

Port not listening – server returns RST immediately.

Application aborts – explicit close() replaced by RST via SO_LINGER.

Stale packets arriving after the connection has been closed.

Capture RST traffic:

tcpdump -i <iface> tcp -w capture.cap

TIME_WAIT and CLOSE_WAIT

Count socket states:

netstat -n | awk '/^tcp/ {++S[$NF]} END {for(a in S) print a, S[a]}'

ss -ant | awk '{++S[$1]} END {for(a in S) print a, S[a]}'

Mitigate excessive TIME_WAIT with kernel parameters:

net.ipv4.tcp_tw_reuse = 1

net.ipv4.tcp_tw_recycle = 1

Reduce tcp_max_tw_buckets to limit total sockets. High CLOSE_WAIT usually means an application thread is stuck after receiving FIN; use jstack to locate the blocked I/O calls.