[EXPORTER] Refactor ElasticSearchRecordable

[sjinks]

Changes

• Simplify ElasticSearchRecordable::WriteValue() by using the ADL Serializer;
• Simplify ElasticSearchRecordable::SetTimestamp();
• Refactor ElasticSearchRecordable::SetTraceId() and ElasticSearchRecordable::SetSpanId() by using constants for array sizes instead of hardcoded values.

For significant contributions please make sure you have completed the following items:

• CHANGELOG.md updated for non-trivial changes
• Unit tests have been added (in another PR)
• Changes in public API reviewed

[netlify]

✅ Deploy Preview for opentelemetry-cpp-api-docs canceled.

Name	Link
🔨 Latest commit	d49d0ec
🔍 Latest deploy log	https://app.netlify.com/sites/opentelemetry-cpp-api-docs/deploys/674e2e19229ea70008a9bf3c

[sjinks]

A possible breaking change here is that the original WriteValue() did not handle spans. If this is important, I can add an if condition to follow the old behavior.

[ShadowMaxLeb]

I just bumped into your changes and I was wondering if this could help putting json objects within the root json ?
The modifications I did was kind of a workaround so we could add keys like user.id that would be visible within ELK like the documentation asks for but the key was still user.id and not an object user under root that would have a key id

My hope would be to be able to be able to create an attribute or resource that contains a json object for some keys:

auto attr = { "user":  R"({"id": 1234}"_json};

[sjinks]

This approach is safer: should kSize change someday, the buffer's size will be automatically updated.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 87.86%. Comparing base (f167c36) to head (d49d0ec).
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3164   +/-   ##
=======================================
  Coverage   87.86%   87.86%           
=======================================
  Files         195      195           
  Lines        6151     6151           
=======================================
  Hits         5404     5404           
  Misses        747      747           

[owent]

Why change it to put_time and ostringstream? In my understandarding, using strftime and snprintf with just one static buffer block will have better performence than using ostringstream and put_time with dynamic memory allocation. And there are also some compatiblity problems with put_time and old compiler's STL.

[sjinks]

A newer version of gcc complained:

warning: ‘__builtin___snprintf_chk’ output may be truncated before the last format character [-Wformat-truncation=]
   47 |     std::snprintf(bufferMilliseconds, sizeof(bufferMilliseconds), ".%06ld",
      |                                                                          ^
In function ‘int snprintf(char*, size_t, const char*, ...)’,
/usr/include/x86_64-linux-gnu/bits/stdio2.h:54:35: note: ‘__builtin___snprintf_chk’ output between 8 and 9 bytes into a destination of size 8
   54 |   return __builtin___snprintf_chk (__s, __n, __USE_FORTIFY_LEVEL - 1,
      |          ~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                                    __glibc_objsize (__s), __fmt,
      |                                    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                                    __va_arg_pack ());
      |                                    ~~~~~~~~~~~~~~~~~

so I tried to come up with something safe that does not involve hardcoded sizes, compiles cleanly, does not have a potential for buffer overflows, and is as fast as std::format("{:%FT%T%Ez}", timePoint).

For example, in

  const static int dateToSecondsSize = 19;
  const static int millisecondsSize  = 8;

it was not immediately evident that millisecondsSize reserves one byte for the null terminator (.123456 is seven bytes, not eight), and it felt like dateSize = dateToSecondsSize + millisecondsSize + timeZoneSize; char bufferDate[dateSize]; could lead to a buffer overflow.

Anyway, I think I have found a faster solution not involving strcat and hardcoded buffer sizes.

[owent]

This can be fixed by declaring bufferMilliseconds as char bufferMilliseconds[millisecondsSize + 1]; . And change the snprinf calling to std::snprintf(bufferMilliseconds, sizeof(bufferMilliseconds) - 1, ".%06ld", static_cast<long>(microseconds.count())); . Thera are a little different between the bebaviour of snprinf in MSVC and GCC.

Anyway, I think I have found a faster solution not involving strcat and hardcoded buffer sizes.

The new way seems better, and thanks.

[owent]

LGTM and thanks.

[marcalff]

LGTM, thanks for the fix.