EOFError: end of buffer reached #474
Comments
|
I managed to grab a copy of a bad
Hope this helps |
|
I work with @kuahyeow and reproduced the same issue on my end. In our environment something appears to fail while loading ---
table_name: ci_builds
classes:
- Ci::Bridge
- Ci::Build
- Ci::Processable
- CommitStatus
- GenericCommitStatus
feature_categories:
- continuous_integration
description: TODO
introduced_by_url: https://gitlab.com/gitlab-org/gitlab/-/commit/046b28312704f3131e72dcd2dbdacc5264d4aa62
milestone: '8.0'
gitlab_schema: gitlab_ci
For some reason Bootsnap is just trying to load a single byte: [201, 210] in /Users/stanhu/.asdf/installs/ruby/3.2.3/lib/ruby/gems/3.2.0/gems/bootsnap-1.18.1/lib/bootsnap/compile_cache/yaml.rb
201: def storage_to_output(data, kwargs)
202: if kwargs&.key?(:symbolize_names)
203: kwargs[:symbolize_keys] = kwargs.delete(:symbolize_names)
204: end
205:
=> 206: unpacker = CompileCache::YAML.msgpack_factory.unpacker(kwargs)
207: unpacker.feed(data)
208: safe_loaded = unpacker.unpack
209: if safe_loaded
210: unpacker.unpack
(byebug) data
"\xC3"I added some debug log messages, and I see that the right file is being loaded. However, |
|
I'm also running into #471. It does seem like the header is getting corrupted somehow because you can see in the hexdump the length is indeed set to 1: It appears the first time the YAML cache is compiled it is correct: However, for some reason the next time I ran |
|
Thanks. Yes very likely the same root cause than #471. The revalidation must corrupt the cache entry somehow. If more people are running into it, I think I'll cut another patch release to disable revalidation (or put it behind a flag). |
|
I see that bootsnap/ext/bootsnap/bootsnap.c Line 663 in 4b6d40e However, when |
|
Interesting catch. If that's indeed it, it shouldn't be too hard to produce a test case. 👀 BTW: I just pushed a |
|
Thanks. This diff fixes the issue for me: diff --git a/ext/bootsnap/bootsnap.c b/ext/bootsnap/bootsnap.c
index 326aade..5a7ddb0 100644
--- a/ext/bootsnap/bootsnap.c
+++ b/ext/bootsnap/bootsnap.c
@@ -115,7 +115,7 @@ static void bs_cache_path(const char * cachedir, const VALUE path, char (* cache
static int bs_read_key(int fd, struct bs_cache_key * key);
static enum cache_status cache_key_equal_fast_path(struct bs_cache_key * k1, struct bs_cache_key * k2);
static int cache_key_equal_slow_path(struct bs_cache_key * current_key, struct bs_cache_key * cached_key, const VALUE input_data);
-static int update_cache_key(struct bs_cache_key *current_key, int cache_fd, const char ** errno_provenance);
+static int update_cache_key(struct bs_cache_key *current_key, uint64_t data_size, int cache_fd, const char ** errno_provenance);
static void bs_cache_key_digest(struct bs_cache_key * key, const VALUE input_data);
static VALUE bs_fetch(char * path, VALUE path_v, char * cache_path, VALUE handler, VALUE args);
@@ -331,9 +331,10 @@ static int cache_key_equal_slow_path(struct bs_cache_key *current_key,
return current_key->digest == cached_key->digest;
}
-static int update_cache_key(struct bs_cache_key *current_key, int cache_fd, const char ** errno_provenance)
+static int update_cache_key(struct bs_cache_key *current_key, uint64_t data_size, int cache_fd, const char ** errno_provenance)
{
lseek(cache_fd, 0, SEEK_SET);
+ current_key->data_size = data_size;
ssize_t nwrite = write(cache_fd, current_key, KEY_SIZE);
if (nwrite < 0) {
*errno_provenance = "update_cache_key:write";
@@ -803,7 +804,7 @@ bs_fetch(char * path, VALUE path_v, char * cache_path, VALUE handler, VALUE args
valid_cache = cache_key_equal_slow_path(¤t_key, &cached_key, input_data);
if (valid_cache) {
if (!readonly) {
- if (update_cache_key(¤t_key, cache_fd, &errno_provenance)) {
+ if (update_cache_key(¤t_key, RSTRING_LEN(input_data), cache_fd, &errno_provenance)) {
exception_message = path_v;
goto fail_errno;
}
@@ -965,7 +966,7 @@ bs_precompile(char * path, VALUE path_v, char * cache_path, VALUE handler)
}
valid_cache = cache_key_equal_slow_path(¤t_key, &cached_key, input_data);
if (valid_cache) {
- if (update_cache_key(¤t_key, cache_fd, &errno_provenance)) {
+ if (update_cache_key(¤t_key, RSTRING_LEN(input_data), cache_fd, &errno_provenance)) {
goto fail;
}
}I wonder if it's simpler to do: current_key->data_size = current_key->size;UPDATE: Probably not, since |
|
Hmm, maybe this is not exactly right. The |
|
Yup, I have what I think is a cleaner patch, but you were right on the money, I'm just writing a proper test case and will open a PR in a few minutes. Thanks a lot, that was a great catch. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I made sure the issue is in bootsnap
After disabling bootsnap, the issue now longer appears
Steps to reproduce
Unfortunately, this only happens reliably with
gdk update(GitLab tooling). I am working to find a simpler reproduction in https://gitlab.com/gitlab-org/gitlab/-/merge_requests/143230#note_1750589800Expected behavior
Actual behavior
The previous version of bootsnap we used as 1.17.0. After the upgrade to 1.18.1,
bundle exec rails db:migratestarted failing with:If I run it with
--backtrace, I notice this lines at the top:I suspect something in 1.18.1 has changed relating to
YAML.load_file. Looking at how Bootsnap monkey patches YAML, it's possible that changes to the bs_fetch function in 1.18.0 might be relate to this bug. df6267fSystem configuration
Bootsnap version: 1.18.1
Ruby version:
ruby 3.2.3 (2024-01-18 revision 52bb2ac0a6) [arm64-darwin22]Rails version: 7.0.8
The text was updated successfully, but these errors were encountered: