'No route to host?' with active Internet connection

[Filazapovich]

Part of people (looks like ~10%) whose download our game is affected by strange error with active and stable Internet connection:

ERROR:SCRIPT: Unable to create HTTP connection to [Some server]. No route to host?

Native gethostbyname works fine and resolves host IP. Logcat from device:

D/libc-netbsd(17157): [getaddrinfo]: mtk hostname=[Some server]; servname=(null); cache_mode=(null), netid=0; mark=0
D/libc-netbsd(17157): getaddrinfo( app_uid:10242
D/libc-netbsd(17157): getaddrinfo() uid prop:
D/libc-netbsd(17157): getaddrinfo() getuid():10242
D/libc-netbsd(17157): [getaddrinfo]: mtk ai_addrlen=0; ai_canonname=(null); ai_flags=0; ai_family=0
D/libc-netbsd(  356): [getaddrinfo]: mtk hostname=[Some server]; servname=(null); cache_mode=local, netid=122; mark=917626
D/libc-netbsd(  356): getaddrinfo( app_uid:0
D/libc-netbsd(  356): [getaddrinfo]: mtk ai_addrlen=0; ai_canonname=(null); ai_flags=0; ai_family=0
D/libc-netbsd(  356): res_queryN name = [Some server], class = 1, type = 28
D/libc-netbsd(  356): res_queryN name = [Some server], class = 1, type = 1
D/libc-netbsd(  356): res_queryN name = [Some server] succeed
D/libc-netbsd(17157): getaddrinfo: [Some server] get result from proxy >>

Thinks about than:

• That is not problem with Internet because browser and other applications works fine.
• http.request uses c-ares lib to resolve host IP. Looks like case with that lib? No DNS Search Domain support in C-Ares calls nodejs/node#17850
• We can partially reproduce this behavior by https://play.google.com/store/apps/details?id=jp.co.taosoftware.android.packetcapture. But some others application doesn't work properly with this VPN.

Our QA have 2 devices with that error: Sony Xperia M5 5603 with Android 5.1 and Lenovo Vibe K5 Plus A6020a46 with Android 5.1.1. That is not case with Android 5.1.

Reproducible with defold 161, 159 and earlier.

---

Build time	2019-09-02T09:40:18.383211
Defold channel	stable
Defold editor sha	45635ad26f85009c52905724e242cc92dd252146
Defold engine sha	45635ad26f85009c52905724e242cc92dd252146
Defold version	1.2.161
GPU	Intel(R) HD Graphics 630
GPU Driver	4.5.0 - Build 23.20.16.4973
Java version	11.0.1+13
OS arch	amd64
OS name	Windows 10
OS version	10.0

[britzl]

Thank you for reporting this. We haven't heard of this being an issue for anyone else but nevertheless we are currently investigating it.

[jhonnyking]

"Lenovo Vibe K5 Plus A6020a46 with Android 5.1.1. That is not case with Android 5.1."
Does this mean that you can't reproduce it on Android 5.1 with that device?
Does the issue only happen on Android, or do you have similar issues on iOS?

I took a look into the link you provided, and i I understand correctly the addition that was made was that you can provide c_ares with a list of search locations that it should look through for the resolv.conf file. This has since been adapted into c_ares, but it seems strange that the default implementation cannot find the resolve file. Could you look into the file system on those devices and see if the /etc/resolv.conf file exists? and possibly if you have read access to it and what its content is.

Edit: The resolv.conf file isn't used on Android. From the c-ares code:

/* Use the Android connectivity manager to get a list

• of DNS servers. As of Android 8 (Oreo) net.dns#
• system properties are no longer available. Google claims this
• improves privacy. Apps now need the ACCESS_NETWORK_STATE
• permission and must use the ConnectivityManager which
• is Java only. */

[Filazapovich]

"Lenovo Vibe K5 Plus A6020a46 with Android 5.1.1. That is not case with Android 5.1."
Does this mean that you can't reproduce it on Android 5.1 with that device?
Does the issue only happen on Android, or do you have similar issues on iOS?

This case is 100% reproducible on Sony Xperia M5 5603 and on Lenovo Vibe K5 Plus A6020a46 from our QA devices bank. Both with Android 5.1 but on other devices with Android 5.1 this error is not reproducible. I have no information about similar issue on iOS.

[jhonnyking]

I was able to run into a few issues with tpacketcapture app. It does fail on android 8+ devices 100% of the time, but only occasionally on the 5.1 devices:

• When it fails for the 5.1 devices, we get ssl handshake failure after 300-1500 sequential requests. If I switch the request to regular http instead, it doesn't seem to fail at all. Yesterday I could do 60k+ sequential requests without any of them failing.
• For the 8+ devices, we don't get any DNS servers at all when running in capture mode, so it's not surprising that it fails. In this case we discussed adding a default DNS server such as 8.8.8.8 and see if it helps.

So for now, until we can patch the DNS case, could you switch your request from https to http and see if it works better? Obviously not ideal, but consider it a workaround for now. I'll see if I can send you a WIP engine that you guys can try out this week.

[Filazapovich]

We made fallback to IP as temporary solution.

[Filazapovich]

I think the better solution of no DNS servers problem is property in game.project [network] section with comma separated list of DNS servers used as default servers if no DNS servers found on system. And default value for this property should be 8.8.8.8. This solution make no problems with China distribution.

[jhonnyking]

I have a PR that sets DNS servers to 8.8.8.8 and 8.8.4.4 (and similar IPV6 variants) if ares didn't find anything from the device configuration, so if you want I could send you that version so you can test it and see if it resolves your issue. I have tested it on a device with android 8 and the tpacketcapture app running and there are no more DNS issues.

If the solution works, then maybe we could add something similar to the game.project, but it needs a round of discussions internally first.

[Filazapovich]

Ok. I'm ready to test.

[jhonnyking]

osx: https://d.defold.com/archive/f9ab130e3b955b5606202b6b883c5e34a32ef560/stable/editor2/Defold-x86_64-darwin.dmg
win32: https://d.defold.com/archive/f9ab130e3b955b5606202b6b883c5e34a32ef560/stable/editor2/Defold-x86_64-win32.zip
linux: https://d.defold.com/archive/f9ab130e3b955b5606202b6b883c5e34a32ef560/stable/editor2/Defold-x86_64-linux.zip

[Filazapovich]

Test failed but without "No route to host?" error.
Code:

local function check_http(url, method)
	http.request(url, method, function(self, id, response)
		print(method .. " " .. url .. " : " .. response.status)
	end, nil, "")
end


local function run_batch_http(method)
	check_http("https://example.com", method)
	check_http("https://graph.facebook.com/me", method)
	check_http("[Some server]", method)
end


local function run_test()
	print("------------------------   run test   -------------------------")
	run_batch_http("GET")
	run_batch_http("POST")
end

Output from Windows and from Android):

DEBUG:SCRIPT: GET [Some server] : 400
DEBUG:SCRIPT: GET https://graph.facebook.com/me : 400
WARNING:DLIB: Unhandled ssl status code: -78 (-FFFFFFB2)
ERROR:SCRIPT: HTTP request to '[Some server]' failed (http result: -1  socket result: -1000)
DEBUG:SCRIPT: POST [Some server] : 0
ERROR:SCRIPT: HTTP request to 'https://example.com' failed (http result: -1  socket result: -5)
DEBUG:SCRIPT: POST https://example.com : 0
DEBUG:SCRIPT: GET https://example.com : 304
ERROR:SCRIPT: HTTP request to 'https://graph.facebook.com/me' failed (http result: -1  socket result: -5)
DEBUG:SCRIPT: POST https://graph.facebook.com/me : 0

[britzl]

Will investigate.

[britzl]

Surprisingly the https request to graph.facebook.com/me fails with a socket error. The -78 is MBEDTLS_ERR_NET_SEND_FAILED. Doing the same with curl gives a response:

{"error":{"message":"An active access token must be used to query information about the current user.","type":"OAuthException","code":2500,"fbtrace_id":"Ai5KpDDPg2Pdeu-APCy2kPr"}}

We must be missing something in our config or in the way we set up timeouts

[britzl]

Correction and some additional observations:

The socket error doesn't show up 100% of the time (more like 50%).

The test provided in the example makes both GET and POST requests at the same time. If I only do GET requests the problem never happens.

If I do only a POST request the problem happens.

[britzl]

Isolating the problem further:

http.request("https://graph.facebook.com/me", "POST", function(self, id, response)
	print("POST : " .. response.status)
end, nil, "")

Sometime results in:

WARNING:DLIB: Unhandled ssl status code: -78 (-FFFFFFB2)

Aaaand it never happens if the request actually contains POST data:

http.request("https://graph.facebook.com/me", "POST", function(self, id, response)
	print("POST : " .. response.status)
end, nil, "foo=bar")

This makes sense from one point of view, but I guess a POST request should contain data.

[britzl]

We will include a fix for 0 content length POSTs on SSL connections in 1.2.165.

[britzl]

@Filazapovich The fix was included in 1.2.164. Can you please confirm that you no longer have any problems with HTTP(S) requests?

[Filazapovich]

We will update Defold to 1.2.164+ only at January.