-
Notifications
You must be signed in to change notification settings - Fork 373
vsock: Kata fails to start with use_vsock = true on ppc64le #1943
Comments
Not sure if this is the issue but from here #947
|
On which kernel are you running? Have a look to this #1203 . It's not an endianess bug in any case. |
I saw it from your kata-collect-data. I guess you're using a buggy kernel. Could you try to upgrade it? As suggested in #1203 (comment) |
@alicefr: I tried with an updated kernel and the results are the same
|
Looks like kata-runtime cannot get a unique context id: cc @devimc
|
I think it's important to understand if it's a kata or kernel bug. We (s390x) also need to check if vsock is correctly working (cc @jschintag ). But as far as I got from @nitkon he has another problem than the bug reported at #1203 (comment) , right? You couldn't set the CID. |
@alicefr Our quick test across all the other architectures seem to show that it works fine with Kata. This issue came up when we tried to move vsock creation to an earlier time on Kata. #1014 (comment) has the info the hang on the host. The code in question that we think is hanging is
|
So as far as I understand he's getting a operation not permitted (errno=1) when he tried to set the CID with ioctl call (from #1943 (comment)). The best would be to have a look into the kernel code and check why it gives back this error. My test is simply replicating what kata is doing at
|
Woah!! Seems like this has been open for a while now and you have chased down & closed quite a bit of these. Thank you very much. 💯 I am now hopefully all caught up and educated. Will wait and see what the kernel debug shows. that ought to help us on next steps. |
@amshinde @alicefr : ioctl:1 error like we had seen earlier. But this error take an hour or so to log in the journalctl, after
|
can you try to run kata standalone with |
@alicefr : Thnx for inputs. Here is the output:
|
Update: After an hour or so,
|
/cc @stefanha just in case something springs to mind.. |
@nitkon it sounds to me that after 1 hour there's a timeout and you get back this error. Can you try this small test program in C. It makes the same test as the test64/32.go program #include <stdio.h>
#include <linux/vhost.h>
#include <errno.h>
#include <sys/ioctl.h>
#include <unistd.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdint.h>
int main() {
uint64_t cid = 0x3;
int rc;
int fd;
if ((fd = open("/dev/vhost-vsock", O_RDWR, 0666)) < 0) {
printf("Error in getting fd errno=%d\n", errno);
}
rc = ioctl(fd, VHOST_VSOCK_SET_GUEST_CID, (void*) &cid);
if (rc != 0){
printf("FAIL:VHOST_VSOCK_SET_GUEST_CID=%lx fd=%d cid=%lx ERRNO=%d\n",VHOST_VSOCK_SET_GUEST_CID, fd, cid, errno);
close(fd);
return 1;
}
close(fd);
printf("SUCCESS:fd=%d cid=%lx \n", fd, cid);
return 0;
} If it fails I'd suggest that you report it to the Power KVM team |
Thanks, @alicefr. The program returned success.
|
now I don't get it anymore. I expected this test to fail |
@alicefr : Changed the VHOST_VSOCK_SET_GUEST_CID to
|
I think this is what I am remembering: https://lkml.org/lkml/2019/1/8/300 |
@grahamwhaley nitkon found that the VHOST_VSOCK_SET_GUEST_CID ioctl constant is incorrect on ppc64le. This makes sense because the kernel code path for that ioctl never returns EPERM, so the code must be executing a different ioctl. Needs to be fixed runtime/virtcontainers/utils/utils_linux.go. |
After fixing the
More debug logs: https://paste.fedoraproject.org/paste/6wxqscai8ZgJM45xnap1FQ |
Hey @pmorjan @stefanha. FYI The test application that is mentioned here, errors out with
Looks like @Pennyzct had hit it too and kernel upgrade had fixed the issue. Not in my case though. |
Possible reason for context deadline exceeded/ VM shutting down from kata-runtime logs post this PR Merge.
source=virtcontainers subsystem=kata_agent vmconsole="initcall virtio_vsock_init+0x0/0xb8 returned -16 after 13 usecs Coming from here? |
I rebuilt the kernel with the following config change from upstream kernel config for 4.19.
But no luck. Getting the same error as mentioned above. |
With standalone qemu command, vsock works on Power :
Guest:
Host:
Where nc-vsock is this However fails with Kata. Here are the runtime & qemu logs.
Complete logs: |
@nitkon - have you tried kernel tracing (ftrace?) to see if that gives any more clues as to why you get the error? |
@jodh-intel : I will try kernel tracing with ftrace. Meanwhile, I tried what was mentioned here and got ...
Does this mean the VM boots fine but the agent, which is out
|
Hi @nitkon - I don't think that shows enough detail to answer your question. The agent (PID 1) clearly starts as you see the announce message but it's unclear if the agent hangs / blocks or dies. You could try enabling developer mode since that will allow the agent to coredump, which although you cannot inspect via the debug console (as you can only run with agent-as-init) may mean you see a kernel log message detecting the crash. |
Finally got it working with the following changes to kernel config and runtime... phew!
|
This is stale issue and already fixed upstream. Closing this |
This updates grpc-go vendor package to v1.11.3 release, to fix server.Stop() handling so that server.Serve() does not wait blindly. Full commit list: d11072e (tag: v1.11.3) Change version to 1.11.3 d06e756 clientconn: add support for unix network in DialContext. (kata-containers#1883) 452c2a7 Change version to 1.11.3-dev d89cded (tag: v1.11.2) Change version to 1.11.2 98ac976 server: add grpc.Method function for extracting method from context (kata-containers#1961) 0f5fa28 Change version to 1.11.2-dev 1e2570b (tag: v1.11.1) Change version to 1.11.1 d28faca client: Fix race when using both client-side default CallOptions and per-call CallOptions (kata-containers#1948) 48b7669 Change version to 1.11.1-dev afc05b9 (tag: v1.11.0) Change version to 1.11.0 f2620c3 resolver: keep full unparsed target string if scheme in parsed target is not registered (kata-containers#1943) 9d2250f status: rename Status to GRPCStatus to avoid name conflicts (kata-containers#1944) 2756956 status: Allow external packages to produce status-compatible errors (kata-containers#1927) 0ff1b76 routeguide: reimplement distance calculation dfbefc6 service reflection can lookup enum, enum val, oneof, and field symbols (kata-containers#1910) 32d9ffa Documentation: Fix broken link in rpc-errors.md (kata-containers#1935) d5126f9 Correct Go 1.6 support policy (kata-containers#1934) 5415d18 Add documentation and example of adding details to errors (kata-containers#1915) 57640c0 Allow storing alternate transport.ServerStream implementations in context (kata-containers#1904) 031ee13 Fix Test: Update the deadline since small deadlines are prone to flakes on Travis. (kata-containers#1932) 2249df6 gzip: Add ability to set compression level (kata-containers#1891) 8124abf credentials/alts: Remove the enable_untrusted_alts flag (kata-containers#1931) b96718f metadata: Fix bug where AppendToOutgoingContext could modify another context's metadata (kata-containers#1930) 738eb6b fix minor typos and remove grpc.Codec related code in TestInterceptorCanAccessCallOptions (kata-containers#1929) 211a7b7 credentials/alts: Update ALTS "New" APIs (kata-containers#1921) fa28bef client: export types implementing CallOptions for access by interceptors (kata-containers#1902) ec9275b travis: add Go 1.10 and run vet there instead of 1.9 (kata-containers#1913) 13975c0 stream: split per-attempt data from clientStream (kata-containers#1900) 2c2d834 stats: add BeginTime to stats.End (kata-containers#1907) 3a9e1ba Reset ping strike counter right before sending out data. (kata-containers#1905) 90dca43 resolver: always fall back to default resolver when target does not follow URI scheme (kata-containers#1889) 9aba044 server: Convert all non-status errors to codes.Unknown (kata-containers#1881) efcc755 credentials/alts: change ALTS protos to match the golden version (kata-containers#1908) 0843fd0 credentials/alts: fix infinite recursion bug [in custom error type] (kata-containers#1906) 207e276 Fix test race: Atomically access minConnecTimout in testing environment. (kata-containers#1897) 3ae2a61 interop: Add use_alts flag to client and server binaries (kata-containers#1896) 5190b06 ALTS: Simplify "New" APIs (kata-containers#1895) 7c5299d Fix flaky test: TestCloseConnectionWhenServerPrefaceNotReceived (kata-containers#1870) f0a1202 examples: Replace context.Background with context.WithTimeout (kata-containers#1877) a1de3b2 alts: Change ALTS proto package name (kata-containers#1886) 2e7e633 Add ALTS code (kata-containers#1865) 583a630 Expunge error codes that shouldn't be returned from library (kata-containers#1875) 2759199 Small spelling fixes (unknow -> unknown) (kata-containers#1868) 12da026 clientconn: fix a typo in GetMethodConfig documentation (kata-containers#1867) dfa1834 Change version to 1.11.0-dev (kata-containers#1863) 46fd263 benchmarks: add flag to benchmain to use bufconn instead of network (kata-containers#1837) 3926816 addrConn: Report underlying connection error in RPC error (kata-containers#1855) 445b728 Fix data race in TestServerGoAwayPendingRPC (kata-containers#1862) e014063 addrConn: keep retrying even on non-temporary errors (kata-containers#1856) 484b3eb transport: fix race causing flow control discrepancy when sending messages over server limit (kata-containers#1859) 6c48c7f interop test: Expect io.EOF from stream.Send() (kata-containers#1858) 08d6261 metadata: provide AppendToOutgoingContext interface (kata-containers#1794) d50734d Add status.Convert convenience function (kata-containers#1848) 365770f streams: Stop cleaning up after orphaned streams (kata-containers#1854) 7646b53 transport: support stats.Handler in serverHandlerTransport (kata-containers#1840) 104054a Fix connection drain error message (kata-containers#1844) d09ec43 Implement unary functionality using streams (kata-containers#1835) 37346e3 Revert "Add WithResolverUserOptions for custom resolver build options" (kata-containers#1839) 424e3e9 Stream: do not cancel ctx created with service config timeout (kata-containers#1838) f9628db Fix lint error and typo (kata-containers#1843) 0bd008f stats: Fix bug causing trailers-only responses to be reported as headers (kata-containers#1817) 5769e02 transport: remove unnecessary rstReceived (kata-containers#1834) 0848a09 transport: remove redundant check of stream state in Write (kata-containers#1833) c22018a client: send RST_STREAM on client-side errors to prevent server from blocking (kata-containers#1823) 82e9f61 Use keyed fields for struct initializers (kata-containers#1829) 5ba054b encoding: Introduce new method for registering and choosing codecs (kata-containers#1813) 4f7a2c7 compare atomic and mutex performance in case of contention. (kata-containers#1788) b71aced transport: Fix a data race when headers are received while the stream is being closed (kata-containers#1814) 46bef23 Write should fail when the stream was done but context wasn't cancelled. (kata-containers#1792) 10598f3 Explain target format in DialContext's documentation (kata-containers#1785) 08b7bd3 gzip: add Name const to avoid typos in usage (kata-containers#1804) 8b02d69 remove .please-update (kata-containers#1800) 1cd2346 Documentation: update broken wire.html link in metadata package. (kata-containers#1791) 6913ad5 Document that all errors from RPCs are status errors (kata-containers#1782) 8a8ac82 update const order (kata-containers#1770) e975017 Don't set reconnect parameters when the server has already responded. (kata-containers#1779) 7aea499 credentials: return Unavailable instead of Internal for per-RPC creds errors (kata-containers#1776) c998149 Avoid copying headers/trailers in unary RPCs unless requested by CallOptions (kata-containers#1775) 8246210 Update version to 1.10.0-dev (kata-containers#1777) 17c6e90 compare atomic and mutex performance for incrementing/storing one variable (kata-containers#1757) 65c901e Fix flakey test. (kata-containers#1771) 7f2472b grpclb: Remove duplicate init() (kata-containers#1764) 09fc336 server: fix bug preventing Serve from exiting when Listener is closed (kata-containers#1765) 035eb47 Fix TestGracefulStop flakiness (kata-containers#1767) 2720857 server: fix race between GracefulStop and new incoming connections (kata-containers#1745) 0547980 Notify parent ClientConn to re-resolve in grpclb (kata-containers#1699) e6549e6 Add dial option to set balancer (kata-containers#1697) 6610f9a Fix test: Data race while resetting global var. (kata-containers#1748) f4b5237 status: add Code convenience function (kata-containers#1754) 47bddd7 vet: run golint on _string files (kata-containers#1749) 45088c2 examples: fix concurrent map accesses in route_guide server (kata-containers#1752) 4e393e0 grpc: fix deprecation comments to conform to standard (kata-containers#1691) 0b24825 Adjust keepalive paramenters in the test such that scheduling delays don't cause false failures too often. (kata-containers#1730) f9390a7 fix typo (kata-containers#1746) 6ef45d3 fix stats flaky test (kata-containers#1740) 98b17f2 relocate check for shutdown in ac.tearDown() (kata-containers#1723) 5ff10c3 fix flaky TestPickfirstOneAddressRemoval (kata-containers#1731) 2625f03 bufconn: allow readers to receive data after writers close (kata-containers#1739) b0e0950 After sending second goaway close conn if idle. (kata-containers#1736) b8cf13e Make sure all goroutines have ended before restoring global vars. (kata-containers#1732) 4742c42 client: fix race between server response and stream context cancellation (kata-containers#1729) 8fba5fc In gracefull stop close server transport only after flushing status of the last stream. (kata-containers#1734) d1fc8fa Deflake tests that rely on Stop() then Dial() not reconnecting (kata-containers#1728) dba60db Switch balancer to grpclb when at least one address is grpclb address (kata-containers#1692) ca1b23b Update CONTRIBUTING.md to CNCF CLA 2941ee1 codes: Add UnmarshalJSON support to Code type (kata-containers#1720) ec61302 naming: Fix build constraints for go1.6 and go1.7 (kata-containers#1718) b8191e5 remove stringer and go generate (kata-containers#1715) ff1be3f Add WithResolverUserOptions for custom resolver build options (kata-containers#1711) 580defa Fix grpc basics link in route_guide example (kata-containers#1713) b7dc71e Optimize codes.String() method using a switch instead of a slice of indexes (kata-containers#1712) 1fc873d Disable ccBalancerWrapper when it is closed (kata-containers#1698) bf35f1b Refactor roundrobin to support custom picker (kata-containers#1707) 4308342 Change parseTimeout to not handle non-second durations (kata-containers#1706) be07790 make load balancing policy name string case-insensitive (kata-containers#1708) cd563b8 protoCodec: avoid buffer allocations if proto.Marshaler/Unmarshaler (kata-containers#1689) 61c6740 Add comments to ClientConn/SubConn interfaces to indicate new methods may be added (kata-containers#1680) ddbb27e client: backoff before reconnecting if an HTTP2 server preface was not received (kata-containers#1648) a4bf341 use the request context with net/http handler (kata-containers#1696) c6b4608 transport: fix race sending RPC status that could lead to a panic (kata-containers#1687) 00383af Fix misleading default resolver scheme comments (kata-containers#1703) a62701e Eliminate data race in ccBalancerWrapper (kata-containers#1688) 1e1a47f Re-resolve target when one connection becomes TransientFailure (kata-containers#1679) 2ef021f New grpclb implementation (kata-containers#1558) 10873b3 Fix panics on balancer and resolver updates (kata-containers#1684) 646f701 Change version to 1.9.0-dev (kata-containers#1682) Fixes: kata-containers#307 Signed-off-by: Peng Tao <bergwolf@gmail.com>
Description of problem
Set
use_vsock = true
in configuration.toml and enabled+CONFIG_BINFMT_MISC=y ; +CONFIG_VSOCKETS=y
in the kata kernel.Expected result
Kata container starts without any issue
Actual result
Fails to start. Hangs/no response after starting the container.
journalctl -f
System:
Output from: kata-collect-data.sh
kata-collect-data.log
The text was updated successfully, but these errors were encountered: