Since 11g Oracle resolves hostnames directly by querying DNS rather than from /etc/hosts, this can introduce delays in establishing a connection to the remote database server.
This could be observed in the strace for a SQL*Plus connection:-
open("/etc/nsswitch.conf", O_RDONLY|0x800, 0666) = 5 ioctl(5, TCGETA, 0x9fffffffffffa360) ERR#25 ENOTTY read(5, "# \n# / e t c / n s s w i t c ".., 8192) = 92 read(5, 0x60000000001c9058, 8192) = 0 close(5) = 0 open("/usr/lib/hpux64/libnss_dns.so.1", O_RDONLY|0x800, 0) = 5 fstat(5, 0x9fffffffffff9de0) = 0 pread(5, "7fE L F 0202010101\0\0\0\0\0\0\0".., 1024, 0) = 1024 stat("/usr/lib/hpux64/dpd", 0x9fffffffffff9390) = 0 open("/usr/lib/hpux64/dpd/libnss_dns.so.1.bpd", O_RDONLY|0x800, 0) ERR#2 ENOENT getuid() = 305 (305) getgid() = 303 (303) mmap(NULL, 49440, PROT_READ|PROT_EXEC, MAP_SHARED|MAP_SHLIB, 5, 0) = 0xc00000000b054000 mmap(NULL, 800, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_SHLIB, 5, 65536) = 0x9fffffffbf68d000 close(5) = 0 getuid() = 305 (305) getgid() = 303 (303) getuid() = 305 (305) getgid() = 303 (303) open("/test/local/oracle/11.1.0.7/lib/libdl.so.1", O_RDONLY|0x800, 0) ERR#2 ENOENT open("/test/local/oracle/11.1.0.7/lib32/libdl.so.1", O_RDONLY|0x800, 0) ERR#2 ENOENT getuid() = 305 (305) getgid() = 303 (303) open("/usr/lib/hpux64/libdl.so.1", O_RDONLY|0x800, 0) = 5 fstat(5, 0x9fffffffffff9cf0) = 0 read(5, "7fE L F 0202010101\0\0\0\0\0\0\0".., 64) = 64 close(5) = 0 socket(AF_INET, SOCK_DGRAM, 0) = 5 ioctl(5, SIOCGIFNUM, 0x9fffffffffff9680) = 0 ioctl(5, SIOCGIFCONF, 0x9fffffffffff9690) = 0 socket(AF_INET6, SOCK_DGRAM, 0) = 6 ioctl(6, SIOCGLIFNUM, 0x9fffffffffff9684) = 0 ioctl(6, SIOCGLIFCONF, 0x9fffffffffff96a0) = 0 ioctl(5, SIOCGIFFLAGS, 0x9fffffffffff96b0) = 0 close(5) = 0 close(6) = 0 gettimeofday(0x9fffffffffff7dd0, NULL) = 0 getpid() = 22968 (22967) open("/etc/resolv.conf", O_RDONLY|0x800, 0666) = 5 ioctl(5, TCGETA, 0x9fffffffffff7da0) ERR#25 ENOTTY read(5, "d o m a i n t e s t . c o m \n".., 8192) = 453 read(5, 0x60000000001dddf8, 8192) = 0 close(5) = 0 ... ... ... socket(AF_INET, SOCK_STREAM, 0) = 5 connect(5, 0x60000000001eba50, 16) = 0 getsockname(5, 0x9fffffffffff9da0, 0x9fffffffffff94c0) = 0 getsockopt(5, SOL_SOCKET, SO_SNDBUF, 0x9fffffffffffa000, 0x9fffffffffffa004) = 0 getsockopt(5, SOL_SOCKET, SO_RCVBUF, 0x9fffffffffffa000, 0x9fffffffffffa004) = 0 setsockopt(5, 0x6, TCP_NODELAY, 0x9fffffffffffa00c, 4) = 0
As you can see from the above trace that after reading the nsswitch.conf,
open("/etc/nsswitch.conf", O_RDONLY|0x800, 0666) = 5
library “libnss_dns.so” is loaded,
open("/usr/lib/hpux64/libnss_dns.so.1", O_RDONLY|0x800, 0) = 5
and then /etc/resolv.conf is read and much later the socket for TCP/IP is opened.
open("/etc/resolv.conf", O_RDONLY|0x800, 0666) = 5
In 10g after nsswitch.conf and loading the library /etc/hosts file was read rather than /etc/resolv.conf. The reason is the change in system call, in 10g the system call was gethostbyname() whereas in 11g its getaddrinfo(). And these system calls require different configuration in nsswitch.conf, gethostbyname() require the use of keyword “hosts” while getaddrinfo() the use the keyword “ipnodes”
So after 11g if there is any delay with the DNS the same would be reflected in listener startup times and even with new connections. In order to resolve this we have to make some changes to nsswitch.conf.
Solution:-
Add below line in the /etc/nsswitch.conf:-
ipnodes: files [NOTFOUND=continue] dns
Adding the above line would instruct calls made by getaddrinfo() to first search in /etc/hosts and then if not found, query the DNS server. Thus all the DNS related delays would be avoided.
Hope you like the articles in case of any doubts/queries please feel free to comment below.
Oracle Database Internal Mechanism 