issue: #33285
- move streaming related proto into pkg.
- add v2 message type and change flush message into v2 message.
Signed-off-by: chyezh <chyezh@outlook.com>
issue: #33285
- implement streaming service client.
- implement producing and consuming service client by streaming coord
client and streaming node client.
Signed-off-by: chyezh <chyezh@outlook.com>
issue: #33285
- make message builder and message conversion type safe
- add adaptor type and function to adapt old msgstream msgpack and
message interface
---------
Signed-off-by: chyezh <chyezh@outlook.com>
issue: #33285
- add idAlloc interface
- fix binary unsafe bug for message
- fix service discovery lost when repeated address with different server
id
---------
Signed-off-by: chyezh <chyezh@outlook.com>
issue: #33285
- add two grpc resolver (by session and by streaming coord assignment
service)
- add one grpc balancer (by serverID and roundrobin)
- add lazy conn to avoid block by first service discovery
- add some utility function for streaming service
Signed-off-by: chyezh <chyezh@outlook.com>
Default llvm toolchain version in Ubuntu 20.04 is 10, while Ubuntu 22.04
does not have `clang-tidy-10` or `clang-format-10` by default.
issue: #33142
Signed-off-by: Patrick Weizhi Xu <weizhi.xu@zilliz.com>
Signed-off-by: Yinzuo Jiang <jiangyinzuo@foxmail.com>
Signed-off-by: shaoting-huang [shaoting-huang@zilliz.com]
issue: https://github.com/milvus-io/milvus/issues/32982
# Background
Go 1.21 introduces several improvements and changes over Go 1.20, which
is quite stable now. According to
[Go 1.21 Release Notes](https://tip.golang.org/doc/go1.21), the big
difference of Go 1.21 is enabling Profile-Guided Optimization by
default, which can improve performance by around 2-14%. Here are the
summary steps of PGO:
1. Build Initial Binary (Without PGO)
2. Deploying the Production Environment
3. Run the program and collect Performance Analysis Data (CPU pprof)
4. Analyze the Collected Data and Select a Performance Profile for PGO
5. Place the Performance Analysis File in the Main Package Directory and
Name It default.pgo
6. go build Detects the default.pgo File and Enables PGO
7. Build and Release the Updated Binary (With PGO)
8. Iterate and Repeat the Above Steps
<img width="657" alt="Screenshot 2024-05-14 at 15 57 01"
src="https://github.com/milvus-io/milvus/assets/167743503/b08d4300-0be1-44dc-801f-ce681dabc581">
# What does this PR do
There are three experiments, search benchmark by Zilliz test platform,
search benchmark by open-source
[VectorDBBench](https://github.com/zilliztech/VectorDBBench?tab=readme-ov-file),
and search benchmark with PGO. We do both search benchmarks by Zilliz
test platform and by VectorDBBench to reduce reliance on a single
experimental result. Besides, we validate the performance enhancement
with PGO.
## Search Benchmark Report by Zilliz Test Platform
An upgrade to Go 1.21 was conducted on a Milvus Standalone server,
equipped with 16 CPUs and 64GB of memory. The search performance was
evaluated using a 1 million entry local dataset with an L2 metric type
in a 768-dimensional space. The system was tested for concurrent
searches with 50 concurrent tasks for 1 hour, each with a 20-second
interval. The reason for using one server rather than two servers to
compare is to guarantee the same data source and same segment state
after compaction.
Test Sequence:
1. Go 1.20 Initial Run: Insert data, build index, load index, and
search.
2. Go 1.20 Rebuild: Rebuild the index with the same dataset, load index,
and search.
3. Go 1.21 Load: Upload to Go 1.21 within the server. Then load the
index from the second run, and search.
4. Go 1.21 Rebuild: Rebuild the index with the same dataset, load index,
and search.
Search Metrics:
| Metric | Go 1.20 | Go 1.20 Rebuild Index | Go 1.21 | Go 1.21 Rebuild
Index |
|----------------------------|------------------|-----------------|------------------|-----------------|
| `search requests` | 10,942,683 | 16,131,726 | 16,200,887 | 16,331,052
|
| `search fails` | 0 | 0 | 0 | 0 |
| `search RT_avg` (ms) | 16.44 | 11.15 | 11.11 | 11.02 |
| `search RT_min` (ms) | 1.30 | 1.28 | 1.31 | 1.26 |
| `search RT_max` (ms) | 446.61 | 233.22 | 235.90 | 147.93 |
| `search TP50` (ms) | 11.74 | 10.46 | 10.43 | 10.35 |
| `search TP99` (ms) | 92.30 | 25.76 | 25.36 | 25.23 |
| `search RPS` | 3,039 | 4,481 | 4,500 | 4,536 |
### Key Findings
The benchmark tests reveal that the index build time with Go 1.20 at
340.39 ms and Go 1.21 at 337.60 ms demonstrated negligible performance
variance in index construction. However, Go 1.21 offers slightly better
performance in search operations compared to Go 1.20, with improvements
in handling concurrent tasks and reducing response times.
## Search Benchmark Report By VectorDb Bench
Follow
[VectorDBBench](https://github.com/zilliztech/VectorDBBench?tab=readme-ov-file)
to create a VectorDb Bench test for Go 1.20 and Go 1.21. We test the
search performance with Go 1.20 and Go 1.21 (without PGO) on the Milvus
Standalone system. The tests were conducted using the Cohere dataset
with 1 million entries in a 768-dimensional space, utilizing the COSINE
metric type.
Search Metrics:
Metric | Go 1.20 | Go 1.21 without PGO
-- | -- | --
Load Duration (seconds) | 1195.95 | 976.37
Queries Per Second (QPS) | 841.62 | 875.89
99th Percentile Serial Latency (seconds) | 0.0047 | 0.0076
Recall | 0.9487 | 0.9489
### Key Findings
Go 1.21 indicates faster index loading times and larger search QPS
handling.
## PGO Performance Test
Milvus has already added
[net/http/pprof](https://pkg.go.dev/net/http/pprof) in the metrics. So
we can curl the CPU profile directly by running
`curl -o default.pgo
"http://${MILVUS_SERVER_IP}:${MILVUS_SERVER_PORT}/debug/pprof/profile?seconds=${TIME_SECOND}"`
to collect the profile as the default.pgo during the first search. Then
I build Milvus with PGO and use the same index to run the search again.
The result is as below:
Search Metrics
| Metric | Go 1.21 Without PGO | Go 1.21 With PGO | Change (%) |
|---------------------------------------------|------------------|-----------------|------------|
| `search Requests` | 2,644,583 | 2,837,726 | +7.30% |
| `search Fails` | 0 | 0 | N/A |
| `search RT_avg` (ms) | 11.34 | 10.57 | -6.78% |
| `search RT_min` (ms) | 1.39 | 1.32 | -5.18% |
| `search RT_max` (ms) | 349.72 | 143.72 | -58.91% |
| `search TP50` (ms) | 10.57 | 9.93 | -6.05% |
| `search TP99` (ms) | 26.14 | 24.16 | -7.56% |
| `search RPS` | 4,407 | 4,729 | +7.30% |
### Key Findings
PGO led to a notable enhancement in search performance, particularly in
reducing the maximum response time by 58% and increasing the search QPS
by 7.3%.
### Further Analysis
Generate a diff flame graphs between two CPU profiles by running `go
tool pprof -http=:8000 -diff_base nopgo.pgo pgo.pgo -normalize`
<img width="1894" alt="goprofiling"
src="https://github.com/milvus-io/milvus/assets/167743503/ab9e91eb-95c7-4963-acd9-d1c3c73ee010">
Further insight of HnswIndexNode and Milvus Search Handler
<img width="1906" alt="hnsw"
src="https://github.com/milvus-io/milvus/assets/167743503/a04cf4a0-7c97-4451-b3cf-98afc20a0b05">
<img width="1873" alt="search_handler"
src="https://github.com/milvus-io/milvus/assets/167743503/5f4d3982-18dd-4115-8e76-460f7f534c7f">
After applying PGO to the Milvus server, the CPU utilization of the
faiss::fvec_L2 function has decreased. This optimization significantly
enhances the performance of the
[HnswIndexNode::Search::searchKnn](e0c9c41aa2/src/index/hnsw/hnsw.cc (L203))
method, which is frequently invoked by Knowhere during high-concurrency
searches. As the explanation from Go release notes, the function might
be more aggressively inlined by Go compiler during the second build with
the CPU profiling collected from the first run. As a result, the search
handler efficiency within Milvus DataNode has improved, allowing the
server to process a higher number of search queries per second (QPS).
# Conclusion
The combination of Go 1.21 and PGO has led to substantial enhancements
in search performance for Milvus server, particularly in terms of search
QPS and response times, making it more efficient for handling
high-concurrency search operations.
Signed-off-by: shaoting-huang <shaoting.huang@zilliz.com>
issue #32476
tested on x86_64 and aarch64. I'm not sure what needs to be done on some
exotic architectures.
Signed-off-by: Alexandr Guzhva <alexanderguzhva@gmail.com>
Install openblas using apt or yum in scripts/install_deps.sh, update
documentations and fix some typos related to build and installation.
issue: #33056, #33066
Signed-off-by: Yinzuo Jiang <jiangyinzuo@foxmail.com>
See also #33062
This PR:
- Add `lock.RWMutex` & `lock.Mutex` alias to switch implementation based
on build flags
- When build flags has `test` in it, use `go-deadlock` to detect
possible deadlocks
- Replace all `sync.RWMutex` & `sync.Mutex` in datacoord pkg
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
issue: #29507
Notice that api_testonly.go files should be guarded by compiler tag
`test`, so that production build rules don't compile them and these APIs
don't get misused.
Signed-off-by: yiwangdr <yiwangdr@gmail.com>
This PR mainly improve two items:
1. Target observer should refresh loading status during init time. An
uninitialized loading status blocks search/query. Currently, the target
observer refreshes every 10 seconds, i.e. we'd need to wait for 10s for
no reason. That's also the reason why we constantly see false log
"collection unloaded" upon mixcoord restarts.
2. Delete session when service is stopped. So that the new service
doesn't need to wait for the previous session to expire (~10s).
Item 1 is the major improvement of this PR, which should speed up init
time by 10s.
Item 2 is not a big concern in most cases as coordinators usually shut
down after stop(). In those cases, coordinator restart triggers serverID
change which further triggers an existing logic that deletes expired
session. This PR only fixes rare cases where serverID doesn't change.
integration test:
`go test -tags dynamic -v -coverprofile=profile.out -covermode=atomic
tests/integration/coordrecovery/coord_recovery_test.go -timeout=20m`
Performance after the change:
Average init time of coordinators: 10s
Hardware: M2 Pro
Test setup: 1000 collections with 1000 rows (dim=128) per collection.
issue: #29409
Signed-off-by: yiwangdr <yiwangdr@gmail.com>
currently integration test may timeout if any case run time is above 3
minutes. This duration was hard coded.
This PR change this duration into a customized parameter and could be
passed via test running commands.
---------
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
This PR defines the new import reader interfaces and implement a binlog
reader for import.
issue: https://github.com/milvus-io/milvus/issues/28521
---------
Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
issue: #29494
1. link with install path's libblob-chunk-manager
2. performance of `ShouldBindWith` is better than `ShouldBindBodyWith`
3. the middleware shouldn't read the unrefreshed parameter repeatly
Signed-off-by: PowderLi <min.li@zilliz.com>
issue: #28898
This PR move the `ProxyClientManager` to util package, in case of
reusing it's implementation in querycoord
Signed-off-by: Wei Liu <wei.liu@zilliz.com>
