Adjust sync cycle (#328)

* Adjust the sync_cycle automatically.

* Change the multiplier type to double.

* Adjust the sync_period.

* Fix comment.

* Add the pre-commit

* Run pre-commit with other files.

README.md

@@ -79,7 +79,7 @@ VisualDL provides both Python SDK and C++ SDK in order to fit more use cases.
 
 
 ### Python SDK
-VisualDL now supports both Python 2 and Python 3. 
+VisualDL now supports both Python 2 and Python 3.
 Below is an example of creating a simple Scalar component and inserting data from different timestamps:
 
 ```python
@@ -162,7 +162,7 @@ pip install --upgrade dist/visualdl-*.whl
 ### Run a demo from scratch
 ```
 # vdl_create_scratch_log is a helper commend that creates mock data.
-vdl_create_scratch_log 
+vdl_create_scratch_log
 visualDL --logdir=scratch_log --port=8080
 ```
 that will start a server locally on port 8080, then

demo/keras/TUTORIAL_EN.md

@@ -87,4 +87,4 @@ The histograms of the training parameters is as follows:
 <img width="70%" src="https://github.com/daming-lu/large_files/blob/master/keras_demo_figs/keras_histogram.png?raw=true" />
 </p>
 
-The full demonstration code can be downloaded in [here](https://github.com/PaddlePaddle/VisualDL/blob/develop/demo/keras/keras_mnist_demo.py).
+The full demonstration code can be downloaded in [here](https://github.com/PaddlePaddle/VisualDL/blob/develop/demo/keras/keras_mnist_demo.py).
\ No newline at end of file

demo/mxnet/TUTORIAL_EN.md

@@ -4,7 +4,7 @@ Here we will show you how to use VisualDL in MXNet so that you can visualize the
 We will use the MXNet Convolution Neural Network to train the [MNIST](http://yann.lecun.com/exdb/mnist/) dataset as an example.
 
 ## Install MXNet
-Please install MXNet according to MXNet's [official website](https://mxnet.incubator.apache.org/install/index.html) 
+Please install MXNet according to MXNet's [official website](https://mxnet.incubator.apache.org/install/index.html)
 and verify that the installation is successful.
 
     >>> import mxnet as mx
@@ -58,7 +58,7 @@ lenet_model.fit(train_iter,
 ```
 
 That's all. In the training process of MXNet, our callback function is called to record the accuracy at the end of each training batch.
-The rate of accuracy will continue to rise until more than 95%. 
+The rate of accuracy will continue to rise until more than 95%.
 The following is the accuracy of the two epochs:
 
 <p align=center><img width="50%" src="https://github.com/PaddlePaddle/VisualDL/blob/develop/demo/mxnet/epoch2_small.png?raw=true" /></p>

demo/pytorch/TUTORIAL_EN.md

@@ -167,7 +167,7 @@ for epoch in range(5):  # loop over the dataset multiple times
 print('Finished Training')
 ```
 
-PyTorch support ONNX standard and it can export its model into ONNX. 
+PyTorch support ONNX standard and it can export its model into ONNX.
 PyTorch runs a single round of inference to trace the graph. We use a dummy input to run the model to produce the ONNX model
 
 ```python

docs/quick_start_en.md

@@ -27,10 +27,10 @@ logw = LogWriter("./random_log", sync_cycle=10000)
 ```
 
 The first parameter points to a folder; the second parameter `sync_cycle` specifies out of how memory operations should be
-store the data into hard drive. 
+store the data into hard drive.
 
 ### sync_cycle
-Writing is a heavy operation. Setting `sync_cycle` might slow down your training. 
+Writing is a heavy operation. Setting `sync_cycle` might slow down your training.
 A good starting point is to set the `sync_cycle` to be at least twice the amount of data point your would like to capture.
 
 There are different modes for model training, such as training, validating and testing. All these correspond to `mode' in VisualDL.

visualdl/logic/im.cc

@@ -22,6 +22,15 @@ limitations under the License. */
 
 namespace visualdl {
 
+const int minimun_sync_cycle = 100;
+// Expect sync happens every 15~25 seconds
+const int sync_period = 20;
+const int period_range = 5;
+const double slower_multiplier = 1.4;
+const double faster_multiplier = 0.5;
+
+static time_t last_sync_time = time(NULL);
+
 template <typename T>
 void SimpleWriteSyncGuard<T>::Start() {
   CHECK(data_);
@@ -33,6 +42,23 @@ void SimpleWriteSyncGuard<T>::End() {
   CHECK(data_);
   if (data_->parent()->meta.ToSync()) {
     Sync();
+
+    time_t current_time = time(NULL);
+    time_t interval = current_time - last_sync_time;
+
+    // If last sync happens more than 25 seconds ago, the system needs to make
+    // the sync-up faster
+    if (interval > sync_period + period_range) {
+      data_->parent()->meta.cycle =
+          std::max(long(data_->parent()->meta.cycle * faster_multiplier),
+                   long(minimun_sync_cycle));
+    } else if (interval < sync_period - period_range) {
+      // If the last sync happens less than 15 seconds ago, the system needs to
+      // make the sync-up slower.
+      data_->parent()->meta.cycle = std::min(
+          long(data_->parent()->meta.cycle * slower_multiplier), LONG_MAX);
+    }
+    last_sync_time = current_time;
   }
 }
 

visualdl/storage/storage.cc

@@ -32,9 +32,7 @@ Storage::Storage(const Storage& other)
   dir_ = other.dir_;
 }
 
-Storage::~Storage() {
-    PersistToDisk();
-}
+Storage::~Storage() { PersistToDisk(); }
 
 void Storage::AddMode(const std::string& x) {
   // avoid duplicate modes.
@@ -54,13 +52,9 @@ Tablet Storage::AddTablet(const std::string& x) {
   return Tablet(&(*tablets_)[x], this);
 }
 
-void Storage::SetDir(const std::string& dir) {
-    *dir_ = dir;
-}
+void Storage::SetDir(const std::string& dir) { *dir_ = dir; }
 
-std::string Storage::dir() const {
-    return *dir_;
-}
+std::string Storage::dir() const { return *dir_; }
 
 void Storage::PersistToDisk() { PersistToDisk(*dir_); }
 
@@ -70,27 +64,25 @@ void Storage::PersistToDisk(const std::string& dir) {
 
   fs::SerializeToFile(*data_, meta_path(dir));
   for (auto tag : data_->tags()) {
-    if (modified_tablet_set_.count(tag) > 0){
-        auto it = tablets_->find(tag);
-        CHECK(it != tablets_->end()) << "tag " << tag << " not exist.";
-        fs::SerializeToFile(it->second, tablet_path(dir, tag));
+    if (modified_tablet_set_.count(tag) > 0) {
+      auto it = tablets_->find(tag);
+      CHECK(it != tablets_->end()) << "tag " << tag << " not exist.";
+      fs::SerializeToFile(it->second, tablet_path(dir, tag));
     }
   }
   modified_tablet_set_.clear();
 }
 
-Storage* Storage::parent() {
-    return this;
-}
+Storage* Storage::parent() { return this; }
 
 void Storage::MarkTabletModified(const std::string tag) {
-    modified_tablet_set_.insert(tag);
+  modified_tablet_set_.insert(tag);
 }
 
 void Storage::AddTag(const std::string& x) {
-    *data_->add_tags() = x;
-    WRITE_GUARD
-  }
+  *data_->add_tags() = x;
+  WRITE_GUARD
+}
 
 // StorageReader
 std::vector<std::string> StorageReader::all_tags() {

visualdl/storage/tablet.cc

@@ -18,18 +18,18 @@ limitations under the License. */
 namespace visualdl {
 
 void Tablet::SetTag(const std::string& mode, const std::string& tag) {
-    auto internal_tag = mode + "/" + tag;
-    string::TagEncode(internal_tag);
-    internal_encoded_tag_ = internal_tag;
-    data_->set_tag(internal_tag);
-    WRITE_GUARD
+  auto internal_tag = mode + "/" + tag;
+  string::TagEncode(internal_tag);
+  internal_encoded_tag_ = internal_tag;
+  data_->set_tag(internal_tag);
+  WRITE_GUARD
 }
 
 Record Tablet::AddRecord() {
-    parent()->MarkTabletModified(internal_encoded_tag_);
-    IncTotalRecords();
-    WRITE_GUARD
-    return Record(data_->add_records(), parent());
+  parent()->MarkTabletModified(internal_encoded_tag_);
+  IncTotalRecords();
+  WRITE_GUARD
+  return Record(data_->add_records(), parent());
 }
 
 TabletReader Tablet::reader() { return TabletReader(*data_); }

visualdl/storage/tablet.h

@@ -29,11 +29,12 @@ struct TabletReader;
  * Tablet is a helper for operations on storage::Tablet.
  */
 struct Tablet {
-  enum Type { kScalar = 0, kHistogram = 1, kImage = 2, kUnknown = -1};
+  enum Type { kScalar = 0, kHistogram = 1, kImage = 2, kUnknown = -1 };
 
   DECL_GUARD(Tablet);
 
-  Tablet(storage::Tablet* x, Storage* parent) : data_(x), x_(parent), internal_encoded_tag_("") {}
+  Tablet(storage::Tablet* x, Storage* parent)
+      : data_(x), x_(parent), internal_encoded_tag_("") {}
 
   static Type type(const std::string& name) {
     if (name == "scalar") {