Filter nest/lift is broken on Windows in 2.1.10 and 2.2.0 =>logs corrupted #8137
Comments
|
anyone? :-) |
janlonsetteig
changed the title
janlonsetteig
changed the title
|
Confirmed also broken on version 2.2.0. Last working version is 2.1.9 |
|
@cosmo0920 @nokute78 Let me know if I can provider more information to help out |
|
@cosmo0920 I don't have build environment on windows, but I tested it. It will fail cause of decoder error and v2.1.10: v2.1.9 |
|
@nokute78 Thanks for looking into this and providing a repro Any idea on how we can get it fixed? |
|
+1 facing this issue as well. Could you please fix? Thank you! |
|
+1 Thank you! |
|
Could someone pick this up, please? 🙏 |
|
Note: Removing |
|
I tested out_file to get raw msgpack output.
Output in hex is v2.1.10: |
|
@leonardo-albertovich Could you check this issue ? We modified log_event_encoder issue for Windows. I think it relates this issue. The issue is that static void helper_pack_string_add_prefix(struct flb_log_event_encoder *log_encoder,
struct filter_nest_ctx *ctx,
const char *str,
int len)
{
flb_log_event_encoder_append_body_values(
log_encoder,
FLB_LOG_EVENT_STRING_LENGTH_VALUE(ctx->prefix_len + len),
FLB_LOG_EVENT_STRING_BODY_VALUE(ctx->prefix, ctx->prefix_len),
FLB_LOG_EVENT_STRING_BODY_VALUE(str, len));
}The flb_log_event_encoder patch is to fix following |
|
assigned to @leonardo-albertovich |
|
+1 |
|
+1 Thank you |
|
@leonardo-albertovich @edsiper |
|
Anyone that can pick up this one? |
|
@leonardo-albertovich I think this issue is caused by a type of Following function adds static void helper_pack_string_add_prefix(struct flb_log_event_encoder *log_encoder,
struct filter_nest_ctx *ctx,
const char *str,
int len)
{
flb_log_event_encoder_append_body_values(
log_encoder,
FLB_LOG_EVENT_STRING_LENGTH_VALUE(ctx->prefix_len + len),
FLB_LOG_EVENT_STRING_BODY_VALUE(ctx->prefix, ctx->prefix_len),
FLB_LOG_EVENT_STRING_BODY_VALUE(str, len));
}
This is caused by 10 + sizeof(size_t) * 3 since |
|
@leonardo-albertovich #include <stdio.h>
#include <stdarg.h>
#include <string.h>
#include <stdint.h>
typedef size_t* flb_log_event_encoder_size_t;
#define FLB_LOG_EVENT_STRING_LENGTH_VALUE_TYPE 1
#define FLB_LOG_EVENT_STRING_BODY_VALUE_TYPE 2
#define FLB_LOG_EVENT_STRING_LENGTH_VALUE(length) \
(int) FLB_LOG_EVENT_STRING_LENGTH_VALUE_TYPE, \
(flb_log_event_encoder_size_t) length
#define FLB_LOG_EVENT_STRING_BODY_VALUE(buffer, length) \
(int) FLB_LOG_EVENT_STRING_BODY_VALUE_TYPE, \
(char *) buffer, \
(flb_log_event_encoder_size_t) length
void Test(int loop_cnt, ...)
{
va_list arg;
int type1;
flb_log_event_encoder_size_t size;
char* str;
int i;
va_start(arg, loop_cnt);
for (i = 0; i < loop_cnt; i++) {
printf("i=%d\n", i);
type1 = va_arg(arg, int);
if (type1 != FLB_LOG_EVENT_STRING_LENGTH_VALUE_TYPE && type1 != FLB_LOG_EVENT_STRING_BODY_VALUE_TYPE) {
printf("%d:type is not length or body\n", i);
}
if (type1 == FLB_LOG_EVENT_STRING_LENGTH_VALUE_TYPE) {
size = va_arg(arg, flb_log_event_encoder_size_t);
printf("Len: size=%ld\n", size);
}
else if (type1 == FLB_LOG_EVENT_STRING_BODY_VALUE_TYPE) {
str = va_arg(arg, char*);
printf("Body: str=%s\n", str);
size = va_arg(arg, flb_log_event_encoder_size_t);
printf("Body: size=%ld\n", size);
}
}
va_end(arg);
}
int main(void) {
int cnt = 3;
size_t prefix_len = 10;
size_t len = 3;
Test(cnt,
FLB_LOG_EVENT_STRING_LENGTH_VALUE(prefix_len+len),
FLB_LOG_EVENT_STRING_BODY_VALUE("_map_data.", prefix_len),
FLB_LOG_EVENT_STRING_BODY_VALUE("key", len)
);
return 0;
} |
|
I don't have build environment for Windows but I sent a patch #8454 |
|
@nokute78 Is there a GitHub build for that so that I can pull a container image with the patch for testing? |
|
@janlonsetteig Sorry I don't know. @patrick-stephens Any ideas to get a following patched container image for windows ? |
|
We do not currently build Windows images for PRs (they take an absolute age and Windows runners cost more as well). Potentially you may be able to tweak the PR to include a CI update to build them though. I'll check if I can trigger a build manually as well, there are some additional workflows for that but not used them for a Windows container before... |
|
@nokute78 if we have a branch on this repo (the main OSS one) then we can trigger a build of all images via: https://github.com/fluent/fluent-bit/actions/workflows/build-branch-containers.yaml These will then be build and uploaded with a custom tag for people to pull from. |
|
Or you can trigger the same workflow on your side @nokute78? It'll be a public image under ghcr.io//... but that should not matter. |
|
@patrick-stephens Thank you for information. |
|
@patrick-stephens Action was failed because of OpenSSL. Could you check it ? |
|
@cosmo0920 Thank you for information. I rebased and run the action. |
|
@nokute78 I'm running ghcr.io/nokute78/fluent-bit/test/fix_8137:windows-2019-fix_8137 in one of our clusters now. It seems to fix the issue. I see logs working again now with nest/lift |
|
@nokute78 Any progress in verifying and releasing the fix? |
|
It is in the review queue, once merged it will go into the next release. |
|
Confirmed fixed in latest version |
Bug Report
Describe the bug
After upgrading from Fluent Bit 2.1.9 to 2.1.10 the nest/lift filter is no longer working and corrupts the logs
My case are Windows pods in k8s
Fluent Bit reads the log files and adds Kubernetes information. Then it nest/lift the Kubernetes fields
The result in cr.fluentbit.io/fluent/fluent-bit:windows-2019-2.1.10 would look something like this:
As you can see there are some weird characters in the log after doing nest/lift. And also some weird
117=>98, 101=>114, 110=>101, 116=>101, 115=>46, 108=>97.All this works fine when I revert the container image to cr.fluentbit.io/fluent/fluent-bit:windows-2019-2.1.9
This is how the output looks in 2.1.9
For Linux containers nest/lift works as before after the upgrade. But I need it working for for Windows pods also
Let me know if you need more information to reproduce the issue
The text was updated successfully, but these errors were encountered: