[PATCH] JDK-8228580 DnsClient TCP socket timeout
milan.mimica at gmail.com
Wed Aug 28 09:41:05 UTC 2019
On Wed, 28 Aug 2019 at 08:26, Vyom Tewari26 <vtewar26 at in.ibm.com> wrote:
> Hi Milan,
> If i am reading code correctly then for each retry you are doubling the retryTimeout(int retryTimeout = (timeout * (1 << retry));) which is right retry strategy, So the timeout set to TCP instance is already double with original timeout and it will keep doubling with each retry.
Check out how DnsClient::doUdpQuery() works. It reduces "timeoutLeft"
after each UDP packed received. This ensures doUdpQuery() method does
not block more than "pktTimeout".
I did the same for TCP. Method DnsClient::continueTcpQuery() performs
Socket::read() 3 times in a row. Note that Socket::setSoTimeout()
applies to each read() operation individually. This means it is
possible that each read() blocks for less than the specified timeout
(hence it will not throw), but in total continueTcpQuery() method
could block almost 3 times the value initially set via
> The "continueTcpQuery" method is getting called from other places as well.
Yes, from whatever code path it may have been called, it has the very
same problem with not respecting DnsClient::timeout.
More information about the core-libs-dev