| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
and restoreState()
Also renames PlanExecutor::getStages() to PlanExecutor::getRootStage()
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1978-07-04T21:24:15Z
ObjectId.fromDate is not correct for all dates
There is an under-run for dates less than 1978-07-04T21:24:15Z ('Math.pow(2,28) - 1' seconds)
Added extra test for upper date limit.
Closes #725
Signed-off-by: Benety Goh <benety@mongodb.com>
|
|
|
|
|
|
| |
FlushViewOfFile from running at the same time"
This reverts commit 5352345eda2ffa24a72af1d5bbaccaa32bcb1006
|
| |
|
|
|
|
| |
the ReplicationCoordinatorImpl
|
|
|
|
| |
ReplicationCoordinatorImpl
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
The JS framework uses DBDirectClient, which requires OperationContext in
order to not conflict with locks, which have already been acquired.
This change also makes dbEval not use pooled scopes and create a new scope
instead.
|
|
|
|
| |
all resources
|
|
|
|
| |
This reverts commit 172d4f6f8bd09d0b2b77d9fb3bf280d01cfb72a1.
|
|
|
|
| |
indexes on admin.system.users and admin.system.roles
|
| |
|
| |
|
|
|
|
|
| |
There are no functional changes, other than the file:line that would be
printed on invariant failures.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Both algorithms cover phases 2 and 3 of forground index building. Importantly
they both take as input pre-sorted data that has already been extracted from
the raw documents (phase 1).
OLD ALGORITHM:
Phase 2:
Add all keys to buckets. When a bucket gets full, create a new bucket and
link it as the original bucket's parent pointer. At the end of the stage
we have a linked-list of leaf nodes starting as the left-most corner.
Phase 3:
Walk the linked produced by phase 2, and build the layer above the leaves.
With each bucket, except for the last (right-most), pop* its highest key
and add it to the bucket that will be it's parent. If the current parent
bucket is full a new one is added to higher-layer linked list as in
phase 1. Phase 3 is repeated on the new higher-layer linked list until you
get a layer with a single bucket which is marked as the root of the tree.
* Importantly, popping a key is what sets the nextChild pointer on a
bucket. It is set to what used to be the prevChildNode of the key that is
popped. A major flaw of this algorithm is that because no key is ever
popped from the root, its nextChild pointer is never filled in.
NEW ALGORITHM:
Phase 2:
Add all keys to buckets. When a bucket gets full, pop the highest key
(setting the nextChild pointer of the bucket to the prevChild of the
popped key), add the popped key to a parent bucket, and create a new right
sibling bucket to add the new key to. If the parent bucket is full, this
same operation is performed on the parent and all full ancestors. If we
get to the root and it is full, a new root is created above the current
root. When creating a new right sibling, it is set as its parent's
nextChild as all keys in the right sibling will be higher than all keys
currently in the parent.
Phase 3:
Does nothing. This may go away soon.
Non-exhaustive list of benefits to the new algorthm:
* Generates a valid btree.
* Doesn't require a second pass over all of the buckets in the index. This can
be significantly faster for indexes that don't fit in ram.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
This implements rollback of collection creation or dropping, including
associated indexes.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
New testing certificates generated for MongoDB using x509gen. Test were also updated slightly to accommodate for new subject alternative names and such.
All instances of "sslAllowInvalidCertificates" were removed from src/mongo/shell.
Closes #724
Signed-off-by: Benety Goh <benety@mongodb.com>
|
|
|
|
|
|
| |
Closes #714
Signed-off-by: Benety Goh <benety@mongodb.com>
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
A prototype hang analyzer for MCI integration to help investigate test
timeouts.
1. Script supports taking dumps, and/or dumping a summary of useful
information about a process
2. Script will iterate through a list of interesting processes (mongo,
mongod, and mongos), and run the tools from step 1.
Supports Linux, MacOS X, and Windows.
|
| |
|
|
|
|
| |
is down
|
| |
|
| |
|
|
|
|
|
| |
Signed-off-by: Matt Kangas <matt.kangas@mongodb.com>
Closes #722
|
|
|
|
|
|
| |
explain_large.js spawns 18 shards and runs a query that allocates
300MB+ of memory on each. This causes OOM conditions on MCI's 32-bit
buildvariants (limited to 4GB of addressable memory, no swap).
|
| |
|
|
|
|
|
|
| |
Closes #721
Signed-off-by: Benety Goh <benety@mongodb.com>
|
|
|
|
|
|
| |
Closes #714
Signed-off-by: Benety Goh <benety@mongodb.com>
|
|
|
|
|
|
| |
Drive write size limitations."
This reverts commit 78c068e5d63d77648954252a785b3673d6e314b7.
|
| |
|
|
|
|
| |
election to ReplicaSetConfig.
|
|
|
|
|
|
|
| |
thread.
This patch prevents a network callback that completes after ReplicationExecutor::shutdown() has been
called from being executed two times.
|
|
|
|
|
|
|
| |
to block indefinitely.
This is useful for testing behaviors that should take place if some events occur
while a network operation is outstanding.
|