Fix panic on 32-bit platforms #10515
Conversation
dnephin
added
type/bug
github-actions
bot
added
theme/certificates
sync/atomic must be used with 64-bit aligned fields, and that alignment is difficult to ensure unless the field is the first one in the struct. https://golang.org/pkg/sync/atomic/#pkg-note-BUG.
The hcl decoding apparently uses strconv.ParseInt, which fails to parse a 64bit int. Since hcl v1 is basically EOl, it seems unlikely we'll fix this in hcl. Since this test is only about loading values from config files, the extra large number doesn't seem important. Trim a few zeros from the numbers so that they parse properly on 32bit platforms. Also skip a slow test when -short is used.
dnephin
force-pushed
the
dnephin/fix-arm32-atomic-aligment
branch
from
86a5aff
to
3b04326
Compare
| @@ -5473,7 +5473,7 @@ func TestLoad_FullConfig(t *testing.T) { | |||
| HTTPSPort: 15127, | |||
| HTTPUseCache: false, | |||
| KeyFile: "IEkkwgIA", | |||
| KVMaxValueSize: 1234567800000000, | |||
| KVMaxValueSize: 1234567800, | |||
There was a problem hiding this comment.
I'm curious about why we changed to use a 32 bit value in this test?
There was a problem hiding this comment.
Good question! I left some detail in the commit message for f34d354
The hcl decoding apparently uses strconv.ParseInt, which fails to parse a 64bit int on a 32bit platform. Since hcl v1 is basically EOL, it seems unlikely we'll fix this in hcl. We run these tests with both json and hcl input, and only the hcl input was failing.
Since this test is only about loading values from config files, the extra large number doesn't seem important, so I trimmed the number a bit to fit into 32bit.
|
🍒 If backport labels were added before merging, cherry-picking will start automatically. To retroactively trigger a backport after merging, add backport labels and re-run https://circleci.com/gh/hashicorp/consul/401754. |
…gment Fix panic on 32-bit platforms Conflicts in tlsutil/config.go were resolved by dropping those changes. The issue that was fixed in that file is not in 1.10.x.
Fixes #10509
Best viewed by individual commit.
Fixes 3 places where a
uint64struct field was used withsync/atomic. The fields were not the first field in the struct, and would cause a panic on 32-bit platforms because the other fields caused theuint64fields to be not 64-bit aligned (https://golang.org/pkg/sync/atomic/#pkg-note-BUG). Also updates the comments to be more explicit about the problem.Adds a new
go-test-32bitCI job to catch the problem. It only runs-shorttests so that we don't have to run a bunch of parallel jobs, but this seems to catch our existing problems even without running all tests. Hopefully all use ofsync/atomicis covered by at least one test that runs in under 100ms.TODO: