Repository navigation
Local DNS resolution sometimes fails #28292
Description
- Version: v12.4.0 + v10.16.0 + v10.13.0
- Platform: Darwin hosea 18.6.0 Darwin Kernel Version 18.6.0: Thu Apr 25 23:16:27 PDT 2019; root:xnu-4903.261.4~2/RELEASE_X86_64 x86_64
- Subsystem: dns
When running multiple DNS queries in parallel using promises, the DNS resolution sometimes fails. We only noticed that on macOS so far.
Sample script:
const dns = require("dns");
const util = require("util");
const requests = [];
for (let i = 0; i < 1000; i++) {
requests.push(
util.promisify(
dns.lookup
)('waldbuehne-heessen.de')
);
}
Promise
.all(requests)
.then(() => {
console.log("ok");
})
.catch(error => {
console.log(error);
return error;
});
Output (sometimes):
Error: getaddrinfo ENOTFOUND waldbuehne-heessen.de
at GetAddrInfoReqWrap.onlookup [as oncomplete] (dns.js:60:26) {
errno: 'ENOTFOUND',
code: 'ENOTFOUND',
syscall: 'getaddrinfo',
hostname: 'waldbuehne-heessen.de'
}
During the execution, no DNS queries are sent out (tested using tcpdump), so only the local DNS cache seems to be involved.
Activity
dploeger commentedon Jun 19, 2019
We also wrote a GO program to check this and we couldn't reproduce the error there.
Also, when using "google.com" the error doesn't occur.
Very strange behaviour.
sam-github commentedon Jun 19, 2019
Can you reproduce when using the DNS promises API instead of util.promisify? https://nodejs.org/api/dns.html#dns_dns_promises_api
mscdex commentedon Jun 19, 2019
Does the Go program always send a DNS network request or does it utilize the same OS-level mechanism node/libuv does?
dploeger commentedon Jun 19, 2019
@sam-github I'll check that and keep you posted
@mscdex There weren't any DNS requests (tested with tcpdump again) at least. I don't know any go-specifics however.
dploeger commentedon Jun 21, 2019
@sam-github Yes, same result with the direct promises API.
bnoordhuis commentedon Jun 24, 2019
This issue is almost certainly a duplicate of #5436 (and see #26903 for context on why the error message is lacking.)
I'm going to refer you to #5436 (and make sure you read through it all, it actually covers several issues) but let me know if there is reason to believe your issue is different and should be reopened.
dploeger commentedon Jun 24, 2019
Right, I already saw #5436 and did try all workarounds but the
family:4one (I somehow skipped that and the comments on the issue are really branching out in multiple directions).So it basically boils down the system library (which isn't glibc on macos I presume) does switch randomly between 4 and 6 resolutions? The tested hosts did not have an AAAA record. So I guess, the successes were 4 resolutions while the fails were 6 resolutions.
I'll comment that on #5436 as well.