Hi all,
Some background, we (CloudFoundry) noticed some weird behavior after
upgrading from go1.7.4 to go1.8.1. An http client that was polling 260
servers started showing more timeouts after the upgrade. After
investigating the issue we realized that the following pattern combined
with tls connection is the culprit:
resp, err := client.Get(url)
if err != nil {
panic(err)
}
defer resp.Body.Close()
m := map[string]string{}
err = json.NewDecoder(resp.Body).Decode(&m)
if err != nil {
panic(err)
}
The reason for this change in behavior between go1.7.4 and go1.8.1 is the
fix for issue#17355 <https://github.com/golang/go/issues/17355>. The new
behavior of the chunkedReader is causing the above code to not read the
entire body, i.e. it falls short of reading the last zero length chunk.
This causes the connection not to be reused on subsequent requests.
I created a small demo app <https://play.golang.org/p/p7edRynwDJ>. *note*
the demo app cannot be run on the playground since it listens and makes
network connections.
I honestly don’t know what’s the right way to fix this issue. One one hand
it seems like the reader and json decoder are following the api contract,
i.e. it’s not really a bug. On the other hand, I think that many go users
use the above pattern which causes very unexpected behavior that is hard to
debug.
The only way I can think of to fix this is either using
ioutil.ReadAll(resp.Body) combined with json.Unmarshal, or use
json.NewDecoder() followed by ioutil.ReadAll and discard the result.
What do you all think ?
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
