filebeat/input/{tcp,udp}: relax requirements that proc entries be present when an address is #37714
Conversation
efd6
added
bug
enhancement
Filebeat
botelastic
bot
added
needs_team
💔 Build Failed
Expand to view the summary
Build stats
Pipeline error
❕ Flaky test reportNo test was executed to be analysed. 🤖 GitHub commentsExpand to view the GitHub comments
To re-run your PR in the CI, just comment with:
|
|
/test |
💔 Build Failed
Expand to view the summary
Build stats
Steps errors
Expand to view the steps failures
|
💔 Build Failed
Expand to view the summary
Build stats
Pipeline error
❕ Flaky test reportNo test was executed to be analysed. 🤖 GitHub commentsExpand to view the GitHub comments
To re-run your PR in the CI, just comment with:
|
💔 Build Failed
Expand to view the summary
Build stats
Pipeline error
❕ Flaky test reportNo test was executed to be analysed. 🤖 GitHub commentsExpand to view the GitHub comments
To re-run your PR in the CI, just comment with:
|
💚 Build Succeeded
Expand to view the summary
Build stats
❕ Flaky test reportNo test was executed to be analysed. 🤖 GitHub commentsExpand to view the GitHub comments
To re-run your PR in the CI, just comment with:
|
💚 Build Succeeded
Expand to view the summary
Build stats
Test stats 🧪
💚 Flaky test reportTests succeeded. 🤖 GitHub commentsExpand to view the GitHub comments
To re-run your PR in the CI, just comment with:
|
…sent when an address is
The previous logic required that if an address is present according to
net.LookupIP, then it must be present in the /proc/net entries. This may
not the case when a tcp/udp listener is created without specifying
tcp4/udp4 for an IPv4 host address and there is an expectation of
finding the socket in the /proc/net/{tcp,udp} table. So only complain if
the entry has ever been found and never skip storing a metric even when
there is a legitimate reason to expect its presence — because it has
been seen in the past. This second part is an extension to reduce the
loss of metric data, even if it is only partial.
This was incorrectly claimed to be decimal due to misreading the kernel source.
💚 Build Succeeded
Expand to view the summary
Build stats
Test stats 🧪
💚 Flaky test reportTests succeeded. 🤖 GitHub commentsExpand to view the GitHub comments
To re-run your PR in the CI, just comment with:
|
…sent when an address is (#37714) The previous logic required that if an address is present according to net.LookupIP, then it must be present in the /proc/net entries. This may not the case when a tcp/udp listener is created without specifying tcp4/udp4 for an IPv4 host address and there is an expectation of finding the socket in the /proc/net/{tcp,udp} table. So only complain if the entry has ever been found and never skip storing a metric even when there is a legitimate reason to expect its presence — because it has been seen in the past. This second part is an extension to reduce the loss of metric data, even if it is only partial. Also fix the base of the queue length parsers. This was incorrectly claimed to be decimal due to misreading the kernel source. (cherry picked from commit b09ac16)
…that proc entries be present when an address is (#37739) * filebeat/input/{tcp,udp}: relax requirements that proc entries be present when an address is (#37714) The previous logic required that if an address is present according to net.LookupIP, then it must be present in the /proc/net entries. This may not the case when a tcp/udp listener is created without specifying tcp4/udp4 for an IPv4 host address and there is an expectation of finding the socket in the /proc/net/{tcp,udp} table. So only complain if the entry has ever been found and never skip storing a metric even when there is a legitimate reason to expect its presence — because it has been seen in the past. This second part is an extension to reduce the loss of metric data, even if it is only partial. Also fix the base of the queue length parsers. This was incorrectly claimed to be decimal due to misreading the kernel source. (cherry picked from commit b09ac16) * remove irrelevant changelog entries --------- Co-authored-by: Dan Kortschak <[email protected]>
…sent when an address is (elastic#37714) The previous logic required that if an address is present according to net.LookupIP, then it must be present in the /proc/net entries. This may not the case when a tcp/udp listener is created without specifying tcp4/udp4 for an IPv4 host address and there is an expectation of finding the socket in the /proc/net/{tcp,udp} table. So only complain if the entry has ever been found and never skip storing a metric even when there is a legitimate reason to expect its presence — because it has been seen in the past. This second part is an extension to reduce the loss of metric data, even if it is only partial. Also fix the base of the queue length parsers. This was incorrectly claimed to be decimal due to misreading the kernel source.
Proposed commit message
The previous logic required that if an address is present according to net.LookupIP, then it must be present in the /proc/net entries. This may not the case when a tcp/udp listener is created without specifying tcp4/udp4 for an IPv4 host address and there is an expectation of finding the socket in the /proc/net/{tcp,udp} table. So only complain if the entry has ever been found and never skip storing a metric even when there is a legitimate reason to expect its presence — because it has been seen in the past. This second part is an extension to reduce the loss of metric data, even if it is only partial.
Also fix the parse base of the receive queue which was incorrectly thought to be decimal from an incorrect reading of the kernel source.
Checklist
CHANGELOG.next.asciidocorCHANGELOG-developer.next.asciidoc.Author's Checklist
How to test this PR locally
Related issues
Use cases
Screenshots
Logs