When delete by partition_key, Milvus will generates L0 segments
globally. During L0 Compaction, those L0 segments will touch all
partitions collection wise. Due to the false-positive rate of segment
bloomfilters, L0 compactions will append false deltalogs to completed
irrelevant partitions, which causes *partition deletion amplification.
This PR uses partition_key to set targeted partitionID when producing
deleteMsgs into MsgStreams. This'll narrow down L0 segments scope to
partition level, and remove the false-positive influence
collection-wise.
However, due to DeleteMsg structure, we can only label one partition to
one deleteMsg, so this enhancement fails if user wants to delete over 2
partition_keys in one deletion.
pr: #38231
See also: #34665
Signed-off-by: yangxuan <xuan.yang@zilliz.com>
issue: #38325
pr: #38342
the old impl only to check grant in default db before drop role, which
may cause role be dropped when grant still exist.
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #37679
pr: #37694
pr #36549 introduce the logic error which update current target when
only parts of channel is ready.
This PR fix the logic error and let dist handler keep pull distribution
on querynode until all delegator becomes serviceable.
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #36293#36242
pr: #36549
after qn recover, delegator may be loaded in new node, after all segment
has been loaded, delegator becomes serviceable. but delegator's target
version hasn't been synced, and if search/query comes, delegator will
use wrong target version to filter out a empty segment list, which
caused empty search result.
This pr will block delegator's serviceable status until target version
is synced
---------
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #37289
pr: #37480
cause pr #37116 introduce retry on get shard leader, which make search
won't fail during query node down.
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #37289
pr: #37288
those test case use search to verify replica's status, but if the search
gap is 1s, the node down's effect may be fixed up by balance.
This PR remove the 1 second gap between search operation.
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #35821
pr: #35822
After collection loaded, if we need to increase/decrease collection's
replica, we need to release and load it again.
milvus offers 4 solution to update loaded collection's replica, this PR
aims to dynamic change the replica number without release, and after
replica number changed, milvus will execute load replica or release
replica in async, and the replica loaded status can be checked by
getReplicas API.
Notice that if set too much replicas than querynode can afford，the new
replica won't be loaded successfully until enough querynode joins.
---------
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #34985#35493
pr: #35492
after querynode has been suspended, it's not allow to load
segment/channel on it, which means the node is read only. to be
compatible with resource group design, after query node has been
suspend, we remove it from it's original resource group, make it a read
only query node in replica. then two things will happens:
1. it's original resource group will be lacking of query nodes, query
coord will assign new node to it.
2. querycoord will try to move out all segments/channels after querynode
has been suspended
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #35170
pr: #35169
This PR enable to set load configs in cluster level, such as replicas
and resource groups. then when load collections will use the load
config.
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
Seals the largest growing segment if the total size of growing segments
of each shard exceeds the size threshold(default 4GB). Introducing this
policy can help keep the size of growing segments within a suitable
level, alleviating the pressure on the delegator.
issue: https://github.com/milvus-io/milvus/issues/34554
pr: https://github.com/milvus-io/milvus/pull/34692
---------
Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
issue: #34595
pr: #34596
When consuming insert data on the delegator node, QueryCoord will move
out some sealed segments to manage its memory usage. After the growing
segment gets flushed, some sealed segments from other workers will be
moved back to the delegator node. To avoid the frequent movement of
segments, we estimate the maximum growing row count and preserve a
fixed-size memory in the delegator node.
---------
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
issue: #32995
pr: #33405
To speed up the construction and querying of Bloom filters, we chose a
blocked Bloom filter instead of a basic Bloom filter implementation.
WARN: This PR is compatible with old version bf impl, but if fall back
to old milvus version, it may causes bloom filter deserialize failed.
In single Bloom filter test cases with a capacity of 1,000,000 and a
false positive rate (FPR) of 0.001, the blocked Bloom filter is 5 times
faster than the basic Bloom filter in both querying and construction, at
the cost of a 30% increase in memory usage.
Block BF construct time {"time": "54.128131ms"}
Block BF size {"size": 3021578}
Block BF Test cost {"time": "55.407352ms"}
Basic BF construct time {"time": "210.262183ms"}
Basic BF size {"size": 2396308}
Basic BF Test cost {"time": "192.596229ms"}
In multi Bloom filter test cases with a capacity of 100,000, an FPR of
0.001, and 100 Bloom filters, we reuse the primary key locations for all
Bloom filters to avoid repeated hash computations. As a result, the
blocked Bloom filter is also 5 times faster than the basic Bloom filter
in querying.
Block BF TestLocation cost {"time": "529.97183ms"}
Basic BF TestLocation cost {"time": "3.197430181s"}
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
sthuang
Zhen Ye
aoiasd
wei liu
XuanYang-cn
yihao.dai
jaime
congqixia
Buqian Zheng
Chun Han
smellthemoon
yellow-shine
cai.zhang
wayblink