[SR-10783] compile command failed due to signal 9

[swift-ci]

Previous ID	SR-10783
Radar	rdar://problem/51181514
Original Reporter	tannernelson (JIRA User)
Type	Bug

Environment

The system is a CircleCI box running Ubuntu 16.04 with 2CPU/4096M.

Swift version 5.0.1 (swift-5.0.1-RELEASE)
Target: x86_64-unknown-linux-gnu

Linux 7b81119c92aa 4.15.0-1035-aws #37-Ubuntu SMP Mon Mar 18 16:15:14 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

Additional Detail from JIRA

Votes	1
Component/s	Compiler, Source Tooling
Labels	Bug, CompilerCrash
Assignee	None
Priority	Medium

md5: a83f0d011f237654ef4bad38a340045d

Issue Description:

Attempting to compile this SPM project causes signal 9:

https://github.com/vapor/crypto/tree/e60bc60b74b9b976800a44369e1067180fa59cba

swift build 
...
<unknown>:0: error: unable to execute command: Killed
<unknown>:0: error: compile command failed due to signal 9 (use -v to see invocation)

• Using --disable-index-store fixes the issue

• Building with -c release does not exhibit the issue.

• Running swift build a second time, after the initial failure (without clearing .build), works

[weissi]

CC @nkcsgexi

[weissi]

@belkadan not sure what exactly the 'Compiler Crash' label is for but this isn't a real crash, this is someone killing the process with SIGKILL. Most likely it's the kernel because of out of memory I'd think.

[belkadan]

It's a crash in practice, even if it's not literally an assertion failure or segfault.

[weissi]

cool, that makes sense

[benlangmuir]

I was not able to reproduce in any setup I have at hand. I tried:

• Swift docker images for 5.0.1 bionic and xenial

• I tried again, but with memory limited to 1.5 GB

• macOS, and I watched the memory use; it never seemed to go over a couple hundred MB per job

I think we'll need more information to investigate this, for example:

• what is the specific command that is failing? You could try adding `-v -Xswiftc -v` to the build to see the invocations

• if this is running out of memory, how much memory is the swift process actually using?

[swift-ci]

Comment by tanner0101 (JIRA)

Here's a link to a PR that will trigger the error in ci:
vapor/open-crypto#93

Here's a link to the failed build:
https://circleci.com/gh/vapor/crypto-kit/1364?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link

The command being run:
swift build -v -Xswiftc -v -Xcc -v -Xlinker -v -Xcxx -v

Resources for this machine:
2CPU/4096MB

The end of the output:

clang -cc1 version 7.0.0 based upon LLVM 7.0.0svn default target x86_64-unknown-linux-gnu
ignoring nonexistent directory "//include"
#include "..." search starts here:
#include <...> search starts here:
 /root/project/.build/x86_64-unknown-linux/debug
 /root/project/Sources/CBcrypt/include
 /root/project/Sources/CBase32/include
 /usr/lib/swift
 //usr/local/include
 /usr/lib/swift/clang/include
 //usr/include/x86_64-linux-gnu
 //usr/include
End of search list.
<unknown>:0: error: unable to execute command: Killed
<unknown>:0: error: compile command failed due to signal 9 (use -v to see invocation)
Exited with code 1

FWIW, I think it is something specific to CircleCI. Perhaps some breach of permissions? The only way I could recreate this locally using Docker was to limit memory very low--about 100M.
But if I SSH into the CircleCI container and run `swift build` (clean, no artifacts) it happens every time.

[benlangmuir]

One thing I noticed in the log is "-j36" in the original build command; maybe swiftpm is detecting the number of cores incorrectly? Does it still fail if you pass `-Xswiftc -j2` or `-Xswiftc -j1`?

[weissi]

ah, this is likely a container CPUs vs. host machine CPUs.

[swift-ci]

Comment by Sergej Jaskiewicz (JIRA)

I'm experiencing this exact issue, have you been able to find a workaround tannernelson (JIRA User)?

Right now I'm just rerunning the whole `swift build` command, which is suboptimal.