Synchronize with develop (#242)

* Fix cache flush (#226)

* Fix a thread race issue that may cause memory error when larger than cache max size data is transferred

* Add a test that writes more data than server cache size

* Fix CI run command

* Update nersc.yml (#238)

* Since PDCinit returns a uint64_t, 0 should indicate failure (#233)

Co-authored-by: Jean Luca Bez <jlbez@lbl.gov>

* Check the return value of `PDC_Client_init` in `PDC_init` (#230)

* Check that return value of PDC_Client_init in PDC_init

* Change return to 0

This will make is simpler when merging #233 (comment)

---------

Co-authored-by: Jean Luca Bez <jlbez@lbl.gov>

* Change `printf` to PDC logger (#232)

* Changed all printf to use pdc logger

Also removed large blocks of comments and chanegd the pdc logger
to print the file name, function, and line number.

* Change typo of LOG_INFO to LOG_ERROR

* Correct grammar from fail -> failed

* update grammer succesfully close -> successfully closed

* switch type of LOG_INFO to LOG_ERROR

* Add logging docs and fix some LOG_INFO->LOG_JUST_PRINT

* update clang formatting

---------

Co-authored-by: Jean Luca Bez <jlbez@lbl.gov>

* Malloc correct size for pdc_obj_metadata_pkg (#237)

Co-authored-by: Jean Luca Bez <jlbez@lbl.gov>

* PDCregion_transfer_create validate client buf, local region, and remote regions (#236)

Co-authored-by: Jean Luca Bez <jlbez@lbl.gov>

---------

Co-authored-by: Houjun Tang <htang4@lbl.gov>
Co-authored-by: Noah Lewis <47840925+TheAssembler1@users.noreply.github.com>

Author: 3 people

.github/workflows/nersc.yml

@@ -1,7 +1,7 @@
 name: NERSC
 
 on:
-  pull_request
+  pull_request_target
 
 env:
   PR_NUMBER: ${{ github.event.number }}

.gitlab-ci.yml

@@ -108,7 +108,7 @@ perlmutter-cache-build:
         - module list
         - mkdir -p ${PDC_BUILD_PATH}/perlmutter/cache
         - cd ${PDC_BUILD_PATH}/perlmutter/cache
-        - cmake ../../.. -DBUILD_MPI_TESTING=ON -DBUILD_SHARED_LIBS=ON -DPDC_SERVER_CACHE=ON -DBUILD_TESTING=ON -DCMAKE_INSTALL_PREFIX=$PDC_DIR -DPDC_ENABLE_MPI=ON -DMERCURY_DIR=$MERCURY_DIR -DCMAKE_C_COMPILER=cc -DMPI_RUN_CMD="srun -A ${PDC_PROJECT} --qos=debug --constraint=cpu --tasks-per-node=64" -DCMAKE_INSTALL_PREFIX=${PDC_INSTALL_PATH}/perlmutter/cache
+        - cmake ../../.. -DBUILD_MPI_TESTING=ON -DBUILD_SHARED_LIBS=ON -DPDC_SERVER_CACHE=ON -DBUILD_TESTING=ON -DPDC_SERVER_CACHE_MAX_GB=1 -DCMAKE_INSTALL_PREFIX=$PDC_DIR -DPDC_ENABLE_MPI=ON -DMERCURY_DIR=$MERCURY_DIR -DCMAKE_C_COMPILER=cc -DMPI_RUN_CMD="srun -A ${PDC_PROJECT} --qos=debug --constraint=cpu --tasks-per-node=64" -DCMAKE_INSTALL_PREFIX=${PDC_INSTALL_PATH}/perlmutter/cache
         - make -j
         - make install
     artifacts:

docs/readme.md

@@ -68,7 +68,7 @@
     - Input: 
       + pdc_name is the reference for PDC class. Recommended use "pdc"
     - Output: 
-      + PDC class ID used for future reference.
+      + PDC class ID used for future reference or 0 on failure.
     - All PDC client applications must call PDCinit before using it. This function will setup connections from clients to servers. A valid PDC server must be running.
     - For developers: currently implemented in pdc.c.
   + perr_t PDCclose(pdcid_t pdcid)

docs/source/developer-notes.rst

@@ -2,6 +2,30 @@
 Developer Notes
 ================================
 
++++++++++++++++++++++++++++++++++++++++++++++
+PDC Logging
++++++++++++++++++++++++++++++++++++++++++++++
+
+The following macros are defined in the header file `src/commons/logging/include/pdc_logger.h` and should be used instead of `printf` for logging.
+
+The available macros, which follow a similar usage to `printf`, are:
+
+.. code-block:: C
+    LOG_ERROR
+    LOG_WARNING
+    LOG_INFO
+    LOG_DEBUG
+    LOG_JUST_PRINT
+
+Each of these macros automatically prepends a timestamp and log level prefix to the output:
+
+.. code-block:: C
+    [year-month-day hour:minute:second] [ERROR] 
+    [year-month-day hour:minute:second] [WARNING] 
+    [year-month-day hour:minute:second] [INFO] 
+    [year-month-day hour:minute:second] [DEBUG] 
+
+`LOG_JUST_PRINT` does not print a prefix and outputs the message as-is. This macro is useful when you need to print raw messages without log metadata.
 
 +++++++++++++++++++++++++++++++++++++++++++++
 PDC Server Metadata Management

examples/C_plus_plus_example/H5Timing.cc

@@ -185,9 +185,9 @@ static int record_timer(std::vector<H5Timer> *timer, const char* filename) {
 
 static int output_results() {
 #ifdef PDC_PATCH
-    printf("total PDCwrite calls = %d, PDCread calls = %d\n", timer_class->PDCwrite_count, timer_class->PDCread_count);
+    LOG_INFO("total PDCwrite calls = %d, PDCread calls = %d\n", timer_class->PDCwrite_count, timer_class->PDCread_count);
 #else
-    printf("total H5Dwrite calls = %d, H5Dread calls = %d\n", timer_class->H5Dwrite_count, timer_class->H5Dread_count);
+    LOG_INFO("total H5Dwrite calls = %d, H5Dread calls = %d\n", timer_class->H5Dwrite_count, timer_class->H5Dread_count);
 #endif
     if ( timer_class->dataset_timers->size() ) {
         record_timer(timer_class->dataset_timers, "dataset_write_record.csv");
@@ -211,10 +211,10 @@ int finalize_timers() {
     gettimeofday(&temp_time, NULL);
     timer_class->total_end_time = (temp_time.tv_usec + temp_time.tv_sec * 1000000) + .0;
 #ifdef PDC_PATCH
-    printf("total program time is %lf, PDCstart time = %lf, PDCwait time = %lf\n", (timer_class->total_end_time - timer_class->total_start_time) / 1000000.0, timer_class->PDCstart_time / 1000000.0, timer_class->PDCwait_time / 1000000.0);
+    LOG_INFO("total program time is %lf, PDCstart time = %lf, PDCwait time = %lf\n", (timer_class->total_end_time - timer_class->total_start_time) / 1000000.0, timer_class->PDCstart_time / 1000000.0, timer_class->PDCwait_time / 1000000.0);
 #else
-    printf("total program time is %lf, H5Dwrite time = %lf, H5Dread time = %lf\n", (timer_class->total_end_time - timer_class->total_start_time) / 1000000.0, timer_class->H5Dwrite_time / 1000000.0, timer_class->H5Dread_time / 1000000.0);
-    printf("merge requests time = %lf, wrap requests time = %lf, H5Dclose = %lf\n", timer_class->merge_requests_time / 1000000.0, timer_class->wrap_requests_time / 1000000.0, timer_class->H5Dclose_time / 1000000.0);
+    LOG_INFO("total program time is %lf, H5Dwrite time = %lf, H5Dread time = %lf\n", (timer_class->total_end_time - timer_class->total_start_time) / 1000000.0, timer_class->H5Dwrite_time / 1000000.0, timer_class->H5Dread_time / 1000000.0);
+    LOG_INFO("merge requests time = %lf, wrap requests time = %lf, H5Dclose = %lf\n", timer_class->merge_requests_time / 1000000.0, timer_class->wrap_requests_time / 1000000.0, timer_class->H5Dclose_time / 1000000.0);
 #endif
     free(timer_class);
 

examples/C_plus_plus_example/multidataset_plugin.cc

@@ -276,7 +276,7 @@ int flush_multidatasets() {
     double start_time;
 #endif
     i = 0;
-    //printf("Rank %d number of datasets to be written %d\n", rank, dataset_size);
+    LOG_DEBUG("Rank %d number of datasets to be written %d\n", rank, dataset_size);
 #if ENABLE_MULTIDATASET==1
     #ifdef H5_TIMING_ENABLE
     increment_H5Dwrite();
@@ -319,7 +319,7 @@ int flush_multidatasets() {
 
     free(multi_datasets_temp);
 #else
-    //printf("rank %d has dataset_size %lld\n", rank, (long long int) dataset_size);
+    LOG_DEBUG("rank %d has dataset_size %lld\n", rank, (long long int) dataset_size);
 #ifdef PDC_PATCH
 
     if (cached_objs.size()) {
@@ -362,7 +362,7 @@ int flush_multidatasets() {
             offset_length = new_end[j] - new_start[j];
             pdcid_t reg = PDCregion_create(1, &offset, &offset_length);
             pdcid_t transfer_request_id = PDCregion_transfer_create(ptr, PDC_WRITE, it->second->did, reg, reg);
-            //printf("did = %lu, starting request offset = %lu, size = %lu, j = %d\n", it->second->did, offset, offset_length, j);
+            LOG_DEBUG("did = %lu, starting request offset = %lu, size = %lu, j = %d\n", it->second->did, offset, offset_length, j);
             PDCregion_transfer_start(transfer_request_id);
             PDCregion_transfer_wait(transfer_request_id);
 

examples/C_plus_plus_example/region_transfer_1D_append.cc

@@ -58,14 +58,14 @@ main(int argc, char **argv)
     }
     sprintf(data_name, "data0_%d", rank);
 
-    printf("rank %d C++ example for writing %d lists of arrays with size %d\n", rank, ARRAY_SIZE, DATA_SIZE);
+    LOG_INFO("rank %d C++ example for writing %d lists of arrays with size %d\n", rank, ARRAY_SIZE, DATA_SIZE);
     for ( i = 0; i < ARRAY_SIZE; ++i ) {
         register_multidataset_request_append(data_name, 0, data, sizeof(int) * DATA_SIZE, H5T_NATIVE_CHAR);
     }
 
     sprintf(data_name, "data1_%d", rank);
 
-    printf("rank %d C++ example for writing %d lists of arrays with size %d\n", rank, ARRAY_SIZE, DATA_SIZE);
+    LOG_INFO("rank %d C++ example for writing %d lists of arrays with size %d\n", rank, ARRAY_SIZE, DATA_SIZE);
     for ( i = 0; i < ARRAY_SIZE; ++i ) {
         register_multidataset_request_append(data_name, 0, data, sizeof(int) * DATA_SIZE, H5T_NATIVE_CHAR);
     }