[OTLP/HTTP] Honor Retry-After header when retrying exports by DCchoudhury15 · Pull Request #4186 · open-telemetry/opentelemetry-cpp

DCchoudhury15 · 2026-06-26T06:23:52Z

Changes

The OTLP/HTTP retry logic wasn't respecting the Retry-After response header. If a server returned a 429 or 503 along with Retry-After, the exporter would ignore that value and always use its configured exponential backoff instead.

This PR updates the retry behavior to honor the Retry-After header as defined in RFC 7231 §7.1.3. Both supported formats are handled:

A delay in seconds (for example, Retry-After: 5)
An HTTP-date (for example, Retry-After: Wed, 21 Oct 2015 07:28:00 GMT)

One implementation detail worth noting is that NextRetryTime() is called from the doRetrySessions polling loop after ReleaseResponse() has already cleared the response headers. Because of that, the parsed Retry-After value is cached in retry_after_time_point_ before ReleaseResponse() is called, and NextRetryTime() uses the cached value instead of trying to parse the headers again.

Unit tests are not included yet. I can add them directly to this PR
if the maintainers would like the plan would be to extend the existing
test server fixture with a /retry-after/ endpoint that sets the header,
and add two test cases to ext/test/http/curl_http_test.cc following
the pattern of the existing ExponentialBackoffRetry test: one for
Retry-After: <delay-seconds> and one for Retry-After: <HTTP-date>.
Just let me know and I will push them here.

Known limitation: The HTTP-date parser currently supports only the IMF-fixdate format. The two obsolete date formats (RFC 850 and asctime) are not supported. Since RFC 7231 specifies IMF-fixdate as the required format for senders, this should not be an issue in practice.

CHANGELOG.md updated for non-trivial changes
Unit tests have been added
Changes in public API reviewed

codecov · 2026-06-26T07:14:24Z

Codecov Report

❌ Patch coverage is 26.43678% with 64 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.90%. Comparing base (b1d403d) to head (b5e6cda).

Files with missing lines	Patch %	Lines
api/include/opentelemetry/common/timestamp.h	0.00%	54 Missing ⚠️
ext/src/http/client/curl/http_operation_curl.cc	69.70%	10 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4186      +/-   ##
==========================================
- Coverage   81.16%   80.90%   -0.26%     
==========================================
  Files         444      444              
  Lines       18800    18886      +86     
==========================================
+ Hits        15258    15277      +19     
- Misses       3542     3609      +67

Files with missing lines	Coverage Δ
...lemetry/ext/http/client/curl/http_operation_curl.h	`90.91% <ø> (ø)`
ext/src/http/client/curl/http_operation_curl.cc	`59.28% <69.70%> (+0.54%)`	⬆️
api/include/opentelemetry/common/timestamp.h	`45.10% <0.00%> (-50.73%)`	⬇️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

owent · 2026-06-30T10:37:50Z

+
+bool ParseRetryAfterDelay(std::string value, std::chrono::seconds &delay)
+{
+  value.erase(0, value.find_first_not_of(" \t\r\n"));


We can use common::StringUtil::Trim to trim string.

owent · 2026-06-30T10:48:40Z

+
+  if (std::all_of(value.begin(), value.end(), [](unsigned char c) { return std::isdigit(c); }))
+  {
+    try


Could you please use the codes just like GetTimeoutFromString in sdk/src/common/env_variables.cc to parse duration? The common codes can be moved into api/include/opentelemetry/common/timestamp.h .

And exception can be disabled by -fno-exception or /EH . We should use OPENTELEMETRY_HAVE_EXCEPTIONS to check if exception is enabled.

Good catch. The try/catch around std::stoull is very likely what's triggering the Bazel noexcept CI failure.

I'll replace it with a manual digit-by-digit parsing loop that performs explicit overflow checks, following the same approach used by GetTimeoutFromString in sdk/src/common/env_variables.cc. Where appropriate, I'll keep the exception-related handling guarded with OPENTELEMETRY_HAVE_EXCEPTIONS.

I'll also move the shared parsing logic into api/include/opentelemetry/common/timestamp.h as you suggested so it can be reused across different call sites.

owent · 2026-06-30T10:49:46Z

+
+bool ParseRetryAfterDate(std::string value, std::chrono::system_clock::time_point &date)
+{
+  value.erase(0, value.find_first_not_of(" \t\r\n"));


We can use common::StringUtil::Trim to trim string.

owent · 2026-06-30T10:53:47Z

+  std::tm tm = {};
+  std::istringstream ss(value);
+
+  ss >> std::get_time(&tm, "%a, %d %b %Y %H:%M:%S");


According to https://datatracker.ietf.org/doc/html/rfc7231#section-7.1.1.1 seems support much more formats and timezone setting.

Thanks for catching this. I went back and checked §7.1.1.1 again, and you're right. The RFC defines three valid HTTP-date formats:

IMF-fixdate: Sun, 06 Nov 1994 08:49:37 GMT

RFC 850: Sunday, 06-Nov-94 08:49:37 GMT

asctime: Sun Nov 6 08:49:37 1994

Right now I'm only handling the first one, so I'll add support for the other two as well since the RFC requires recipients to accept all three.

I'll also implement the RFC 850 two-digit year handling. If the parsed year ends up looking more than 50 years in the future, I'll map it back to the most recent matching year in the past, as required by the spec.

For the timezone part, I don't think any extra handling is needed. All three formats are defined as UTC the first two explicitly use GMT, and the asctime format is also specified to be interpreted as UTC. So treating the parsed std::tm as UTC with PortableTimegm should be the right thing to do here.

You're right — all valid timezones in HTTP-date are UTC/GMT, so timezone handling isn't needed. Only the datetime format variations matter.

…-telemetry#4172 Signed-off-by: DCchoudhury15 <divyanshuchoudhury3@gmail.com>

…parsing bugs

pellared · 2026-07-15T17:39:34Z

@DCchoudhury15, there are build failures...

linux-foundation-easycla · 2026-07-18T05:09:45Z

The committers listed above are authorized under a signed CLA.

✅ login: DCchoudhury15 / name: DCchoudhury15 (15b6dd3, 6c9b9bf, cc1e070)
✅ login: DCchoudhury15 / name: Divyanshu Choudhury (127aac6, 760806f)

Use the same decltype(result) overflow-check idiom as GetTimeoutFromString in sdk/src/common/env_variables.cc so the if-condition fits on a single line and matches clang-format output.

Restore the parentheses around std::numeric_limits<...>::max (the (std::numeric_limits<...>::max)() idiom used elsewhere in the codebase) so the call is not expanded as the Windows max macro when timestamp.h is included by translation units that pull in <windows.h>. Keep the decltype(result) form so the line fits within the column limit.

owent · 2026-07-19T15:11:46Z

+    std::chrono::seconds::rep result = 0;
+    for (const char c : value)
+    {
+      if (!std::isdigit(static_cast<unsigned char>(c)))


std::isdigit accepts an int as parameter, and this may cause conversation between signed and unsigned integer in some compilers. The CI doesn't appear to be warning about this. Could you please use if (!std::isdigit(static_cast<int>(static_cast<unsigned char>(c)))) here?

owent · 2026-07-19T15:24:30Z

+    if (line_len > kRetryAfterLen)
+    {
+      bool match = true;
+      for (size_t i = 0; i < kRetryAfterLen && match; ++i)


You can use strncmp instead.

owent · 2026-07-19T15:26:47Z

+
+      if (match)
+      {
+        value = std::string(line + kRetryAfterLen, line_end);


Using a nostd::string_view to output value will have a better performance.

owent · 2026-07-19T15:32:34Z

+    ss.clear();
+    ss.str(str);
+    tm = {};
+    ss >> std::get_time(&tm, "%a %b %d %H:%M:%S %Y");


Do we need ss.imbue(std::locale::classic()); before std::get_time ? Or it may return a name not in English.

owent · 2026-07-19T15:40:57Z

+      std::time_t now_t = std::chrono::system_clock::to_time_t(std::chrono::system_clock::now());
+      int current_year  = 1970 + static_cast<int>(now_t / 31556952);
+      int full_year     = 1900 + tm.tm_year;
+      if (full_year - current_year > 50)


The two-digit year handling relies on std::get_time's %y mapping 69–99 to 19xx and 00–68 to 20xx. This split point comes from POSIX strptime and isn't actually guaranteed by the C++ standard, so it's a non-portable assumption — could you add a comment documenting it?
Also, the RFC 7231 rule (section 7.1.1.1) is that a two-digit year more than 50 years in the future MUST be interpreted as the most recent past year with the same last two digits. The current full_year - current_year > 50 check is equivalent for valid inputs, but normalizing via (full_year - current_year + 50) / 100 would express the intent more directly and be robust to any %y mapping.

DCchoudhury15 requested a review from a team as a code owner June 26, 2026 06:23

owent reviewed Jun 30, 2026

View reviewed changes

Comment thread ext/src/http/client/curl/http_operation_curl.cc Outdated

owent reviewed Jun 30, 2026

View reviewed changes

Comment thread ext/src/http/client/curl/http_operation_curl.cc Outdated

[OTLP/HTTP] Honor Retry-After header when retrying exports Fixes open…

6c9b9bf

…-telemetry#4172 Signed-off-by: DCchoudhury15 <divyanshuchoudhury3@gmail.com>

DCchoudhury15 force-pushed the fix/otlp-http-honor-retry-after branch from 8cdef0a to 6c9b9bf Compare July 6, 2026 08:25

DCchoudhury15 marked this pull request as draft July 6, 2026 08:56

DCchoudhury15 added 2 commits July 9, 2026 08:16

Merge branch 'main' into fix/otlp-http-honor-retry-after

760806f

Move Retry-After parsing into common::HttpUtil, fix delay=0 and date-…

15b6dd3

…parsing bugs

DCchoudhury15 marked this pull request as ready for review July 9, 2026 08:02

Fixing ci build errors

cc1e070

DCchoudhury15 force-pushed the fix/otlp-http-honor-retry-after branch from 8b1ae78 to cc1e070 Compare July 18, 2026 03:58

Merge branch 'main' into fix/otlp-http-honor-retry-after

127aac6

DCchoudhury15 force-pushed the fix/otlp-http-honor-retry-after branch from 5f3fb91 to 127aac6 Compare July 18, 2026 05:14

DCchoudhury15 added 3 commits July 18, 2026 10:46

Fix clang-format violation in timestamp.h ParseDelaySeconds

3aea2ed

Use the same decltype(result) overflow-check idiom as GetTimeoutFromString in sdk/src/common/env_variables.cc so the if-condition fits on a single line and matches clang-format output.

Merge branch 'main' into fix/otlp-http-honor-retry-after

b5e6cda

owent reviewed Jul 19, 2026

View reviewed changes

Uh oh!

Conversation

DCchoudhury15 commented Jun 26, 2026

Changes

Uh oh!

codecov Bot commented Jun 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

pellared commented Jul 15, 2026

Uh oh!

linux-foundation-easycla Bot commented Jul 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

owent Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov Bot commented Jun 26, 2026 •

edited

Loading

linux-foundation-easycla Bot commented Jul 18, 2026 •

edited

Loading

owent Jul 19, 2026 •

edited

Loading