[native] Transfer idle HTTPSession between IO threads in exchange source#22903
Closed
Yuhta wants to merge 1 commit intoprestodb:masterfrom
Closed
[native] Transfer idle HTTPSession between IO threads in exchange source#22903Yuhta wants to merge 1 commit intoprestodb:masterfrom
Yuhta wants to merge 1 commit intoprestodb:masterfrom
Conversation
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
Yuhta
changed the title
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 3, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 3, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
aditi-pandit
previously approved these changes
Jun 3, 2024
Contributor
aditi-pandit
left a comment
aditi-pandit
left a comment
There was a problem hiding this comment.
Thanks @Yuhta. This looks good. We will do a benchmark run to assess how this works in our systems.
xiaoxmeng
reviewed
Jun 4, 2024
presto-native-execution/presto_cpp/main/PeriodicTaskManager.cpp
Outdated
Show resolved
Hide resolved
presto-native-execution/presto_cpp/main/PeriodicTaskManager.cpp
Outdated
Show resolved
Hide resolved
presto-native-execution/presto_cpp/main/PeriodicTaskManager.cpp
Outdated
Show resolved
Hide resolved
presto-native-execution/presto_cpp/main/PeriodicTaskManager.cpp
Outdated
Show resolved
Hide resolved
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 4, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 4, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 4, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 4, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
xiaoxmeng
previously approved these changes
Jun 4, 2024
presto-native-execution/presto_cpp/main/PeriodicTaskManager.cpp
Outdated
Show resolved
Hide resolved
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 4, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
Collaborator
|
This pull request was exported from Phabricator. Differential Revision: D57842433 |
xiaoxmeng
approved these changes
Jun 4, 2024
Contributor
Author
Yuhta
added a commit
to Yuhta/presto
that referenced
this pull request
Jun 5, 2024
… source (prestodb#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
xiaoxmeng
pushed a commit
that referenced
this pull request
Jun 5, 2024
… source (#22903) Summary: Currently each endpoint (server) has only one `SessionPool`, which can only be attached to one `EventBase`. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time. Fix this by creating one `SessionPool` for each endpoint and event base pair. Allow transfer of idle session between different event bases by using `ServerIdleSessionController`. This way we eliminate the bottleneck in event base while still keep the `HttpSession`s reusable. After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (`EVP_DigestSignFinal`) is not visible in shuffle heavy queries when connection pool is enabled. Reviewed By: xiaoxmeng, arhimondr Differential Revision: D57842433
czentgr
added a commit
to czentgr/presto
that referenced
this pull request
Jun 13, 2024
…exchange source (prestodb#22903)" This reverts commit ab3bda7.
czentgr
added a commit
to czentgr/presto
that referenced
this pull request
Jun 17, 2024
…exchange source (prestodb#22903)" This reverts commit ab3bda7.
36 tasks
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Summary:
Currently each endpoint (server) has only one
SessionPool, which can only be attached to oneEventBase. If there is some skewness among the distribution (i.e. too many endpoints being attached to one event base), that event base would become bottleneck and cause regression in query wall time.Fix this by creating one
SessionPoolfor each endpoint and event base pair. Allow transfer of idle session between different event bases by usingServerIdleSessionController. This way we eliminate the bottleneck in event base while still keep theHttpSessions reusable.After the fix, enabling connection pool should no longer cause any regression to wall time. The SSL handshake cost (
EVP_DigestSignFinal) is not visible in shuffle heavy queries when connection pool is enabled.Reviewed By: xiaoxmeng, arhimondr
Differential Revision: D57842433