This repository was archived by the owner on Aug 30, 2022. It is now read-only.
Backport crossdock fixes #7
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Isaac Hier <isaachier@gmail.com>
Signed-off-by: Isaac Hier <isaachier@gmail.com>
Signed-off-by: Isaac Hier <isaachier@gmail.com>
Signed-off-by: Isaac Hier <isaachier@gmail.com>
Signed-off-by: Isaac Hier <isaachier@gmail.com>
black-adder
approved these changes
Nov 14, 2017
| } | ||
|
|
||
| void getSamplingStrategy(SamplingStrategyResponse& result, | ||
| const std::string& serviceName) override | ||
| { | ||
| if (_serverAddr == net::IPAddress()) { |
There was a problem hiding this comment.
hmm, not following. _serverAddr is set to net::IPAddress() if the client couldn't reach agent?
There was a problem hiding this comment.
Yes. net::IPAddress() is the default value for an empty IP and the address is only set upon successful connection. The reason for this is to avoid contacting the server if we couldn't connect on construction. I guess the real solution is not to start the looping thread at all, but that is another class that ends up calling this periodically.
| /build | ||
| /crossdock/crossdock | ||
| /crossdock/jaeger-docker-compose.yml |
There was a problem hiding this comment.
this a temporary thing? don't think we should be ignoring this completely
There was a problem hiding this comment.
crossdock/crossdock is the executable binary itself. Actually changed this in my recent crossdock development. The jaeger-docker-compose.yml is a downloaded file to launch the other jaeger clients.
added 2 commits
November 14, 2017 15:31
Signed-off-by: Isaac Hier <ihier@uber.com>
Signed-off-by: Isaac Hier <ihier@uber.com>
|
I'm closing this in favor of real crossdock pull request |
tudor
added a commit
to rockset/jaeger-client-cpp
that referenced
this pull request
Sep 15, 2018
`Span::isFinished()` returns true iff the span's duration is zero. So, if `steady_clock::now()` returns the same value at span creation as at `Finish()` (unlikely, but possible, especially in virtualized environments), the following scenario can happen: - `Span::FinishWithOptions` sets `_duration` to zero - and then calls `reportSpan`, which calls `RemoteReporter::report` - which acquires the reporter mutex and - adds the span to the reporter's `_queue` Later: - `RemoteReporter::sweepQueue` acquires the reporter mutex - `RemoteReporter::sweepQueue` makes a copy of the span (https://github.com/jaegertracing/jaeger-client-cpp/blob/7e9a135b362c47f7e4d42e0693b05b91a70e6888/src/jaegertracing/reporters/RemoteReporter.cpp#L95) - The span is serialized in `RemoteReporter::sendSpan` - The span (the copy made above) needs to be destroyed because the block in `sweepQueue` ends. - The span's destructor is called, which calls `FinishWithOptions` again - ... which sees that `_duration` is zero, so `isFinished()` returns false: https://github.com/jaegertracing/jaeger-client-cpp/blob/7e9a135b362c47f7e4d42e0693b05b91a70e6888/src/jaegertracing/Span.cpp#L94 - so it tries to call `reportSpan` again - which tries to acquire the reporter's mutex - but we're holding the lock in `sweepQueue`, so - deadlock. Relevant stack trace: ``` (gdb) bt (gdb) bt #0 __lll_lock_wait () at ../sysdeps/unix/sysv/linux/x86_64/lowlevellock.S:135 jaegertracing#1 0x00007fb09c722023 in __GI___pthread_mutex_lock (mutex=0x7fb088b1c1c0) at ../nptl/pthread_mutex_lock.c:78 jaegertracing#2 0x00007fb08860987b in jaegertracing::reporters::RemoteReporter::report(jaegertracing::Span const&) () from /usr/local/lib/libjaegertracing_plugin.so jaegertracing#3 0x00007fb0885b6506 in jaegertracing::Span::FinishWithOptions(opentracing::v2::FinishSpanOptions const&) [clone .localalias.361] () from /usr/local/lib/libjaegertracing_plugin.so jaegertracing#4 0x00007fb08860849a in jaegertracing::Span::~Span() [clone .constprop.253] () from /usr/local/lib/libjaegertracing_plugin.so jaegertracing#5 0x00007fb08860a8fa in jaegertracing::reporters::RemoteReporter::sweepQueue() () from /usr/local/lib/libjaegertracing_plugin.so jaegertracing#6 0x00007fb0886f2f5f in execute_native_thread_routine () from /usr/local/lib/libjaegertracing_plugin.so jaegertracing#7 0x00007fb09c71f6db in start_thread (arg=0x7fb087574700) at pthread_create.c:463 jaegertracing#8 0x00007fb0989ae88f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95 ``` Fixed by enforcing a minimum span duration of 1 `SteadyClock` tick. Signed-off-by: Tudor Bosman <tudor@rockset.com>
isaachier
pushed a commit
that referenced
this pull request
Sep 18, 2018
* Fix deadlock if steady_clock::now() returns the same value twice `Span::isFinished()` returns true iff the span's duration is zero. So, if `steady_clock::now()` returns the same value at span creation as at `Finish()` (unlikely, but possible, especially in virtualized environments), the following scenario can happen: - `Span::FinishWithOptions` sets `_duration` to zero - and then calls `reportSpan`, which calls `RemoteReporter::report` - which acquires the reporter mutex and - adds the span to the reporter's `_queue` Later: - `RemoteReporter::sweepQueue` acquires the reporter mutex - `RemoteReporter::sweepQueue` makes a copy of the span (https://github.com/jaegertracing/jaeger-client-cpp/blob/7e9a135b362c47f7e4d42e0693b05b91a70e6888/src/jaegertracing/reporters/RemoteReporter.cpp#L95) - The span is serialized in `RemoteReporter::sendSpan` - The span (the copy made above) needs to be destroyed because the block in `sweepQueue` ends. - The span's destructor is called, which calls `FinishWithOptions` again - ... which sees that `_duration` is zero, so `isFinished()` returns false: https://github.com/jaegertracing/jaeger-client-cpp/blob/7e9a135b362c47f7e4d42e0693b05b91a70e6888/src/jaegertracing/Span.cpp#L94 - so it tries to call `reportSpan` again - which tries to acquire the reporter's mutex - but we're holding the lock in `sweepQueue`, so - deadlock. Relevant stack trace: ``` (gdb) bt (gdb) bt #0 __lll_lock_wait () at ../sysdeps/unix/sysv/linux/x86_64/lowlevellock.S:135 #1 0x00007fb09c722023 in __GI___pthread_mutex_lock (mutex=0x7fb088b1c1c0) at ../nptl/pthread_mutex_lock.c:78 #2 0x00007fb08860987b in jaegertracing::reporters::RemoteReporter::report(jaegertracing::Span const&) () from /usr/local/lib/libjaegertracing_plugin.so #3 0x00007fb0885b6506 in jaegertracing::Span::FinishWithOptions(opentracing::v2::FinishSpanOptions const&) [clone .localalias.361] () from /usr/local/lib/libjaegertracing_plugin.so #4 0x00007fb08860849a in jaegertracing::Span::~Span() [clone .constprop.253] () from /usr/local/lib/libjaegertracing_plugin.so #5 0x00007fb08860a8fa in jaegertracing::reporters::RemoteReporter::sweepQueue() () from /usr/local/lib/libjaegertracing_plugin.so #6 0x00007fb0886f2f5f in execute_native_thread_routine () from /usr/local/lib/libjaegertracing_plugin.so #7 0x00007fb09c71f6db in start_thread (arg=0x7fb087574700) at pthread_create.c:463 #8 0x00007fb0989ae88f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95 ``` Fixed by enforcing a minimum span duration of 1 `SteadyClock` tick. Signed-off-by: Tudor Bosman <tudor@rockset.com> * fix typo Signed-off-by: Tudor Bosman <tudor@rockset.com>
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.