SPEC-1396 Drivers MUST clear connection pools when SDAM monitoring fails due to a network error #665
Conversation
…ils due to a network error
p-mongo
changed the title
| When a server `check`_ fails due to a network error, | ||
| the client SHOULD clear its connection pool for the server: | ||
| When a server `check`_ fails due to a network error (including a timeout), | ||
| the client MUST clear its connection pool for the server: |
There was a problem hiding this comment.
Any ideas of how we can test this? Maybe in the CMAP spec tests we can configure isMaster to fail once with a network error via configureFailPoint, wait for the monitor to run the doomed isMaster, and then assert that the pool has been cleared.
There was a problem hiding this comment.
I'm not saying we need tests in this PR. Just spitballing. The current changes LGTM.
There was a problem hiding this comment.
I wrote several tests in mongodb/mongo-ruby-driver#1475 but I would prefer to wait until the unified spec runner work is complete before proposing spec tests that touch both cmap and sdam.
There was a problem hiding this comment.
Do we need to wait for that? It's a non-goal in the unified spec runner scope to define the format for non-”operations” tests, including CMAP and SDAM.
There was a problem hiding this comment.
Our cmap spec runner does not do any i/o, thus as it is implemented it is unable to set fail points on the server.
There was a problem hiding this comment.
@ShaneHarvey It appears that failcommand fail point is ignored for ismaster commands.
There was a problem hiding this comment.
Turns out it does work when ismaster is given in mixed case, https://jira.mongodb.org/browse/SERVER-44414.
| When a server `check`_ fails due to a network error, | ||
| the client SHOULD clear its connection pool for the server: | ||
| When a server `check`_ fails due to a network error (including a timeout), | ||
| the client MUST clear its connection pool for the server: |
There was a problem hiding this comment.
I'm not saying we need tests in this PR. Just spitballing. The current changes LGTM.
|
|
||
| When a server `check`_ fails due to a network error, | ||
| the client SHOULD clear its connection pool for the server: | ||
| When a server `check`_ fails due to a network error (including a timeout), |
There was a problem hiding this comment.
The spec defines terms "network error" and "network timeout", should that second term be used here?
| if the monitor's socket is bad it is likely that all are. | ||
| (See `JAVA-1252 <https://jira.mongodb.org/browse/JAVA-1252>`_.) | ||
| (See `JAVA-1252 <https://jira.mongodb.org/browse/JAVA-1252>`_, | ||
| `SPEC-1396 <https://jira.mongodb.org/browse/SPEC-1396>`_.) |
There was a problem hiding this comment.
SPEC tickets are not public, I don't think we can include this in our public documentation
No description provided.