cnvrg 1.11.9 → 1.11.15

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b20aff57e3987256b096469eb8c220743d4d06a70cceaf87d23a81a3c193e6c0
-  data.tar.gz: c70e86b8dd64dc5970b41290db84db83a1f444dbbc83cebf91c3c4c236ca8ffd
+  metadata.gz: 8798ab84db1e0742dc733a0f74b5830e2c5396099fc52bdb4a04ef108a43408e
+  data.tar.gz: 795988ec46ba506091136240bda258f923469e4b3000d58e84371df09f65840d
 SHA512:
-  metadata.gz: 4c6a9bc704fc23a78afd001de3f906751a2b9178bb9ff646c49a4418386933b8b1397b166a4cee1529e1c8332284e79870b2ad847cd4c8df6cf255145bed6e29
-  data.tar.gz: d35b130d510c2595c79da10b189d0ffee424aa211c2033ba49a4fa22fba988cc935d5fb6d20aff4d190cb0de97a30263e9c24ecd6a15251bbce2c6adedc74e46
+  metadata.gz: '0497d165396b52d2975c3d3ab8c097596c173f04c63711519cf2c122fb9dc5f90ecc493aab7ba2cba0d7064dc14b52a7e1d7f2f6f5e68d3af020d4b3f1f5b35c'
+  data.tar.gz: 72c65364f4c9ad8d5acf68546f0a428c113dd9697613674ce83b52a3ce7fcec8064a7517d9b5b5ae7f21a8690ba935d028622688d72f87eb3e1927d7fe5ea9e2

data/lib/cnvrg/api.rb

@@ -65,6 +65,16 @@ module Cnvrg
                     response = conn.get "#{resource}", data
                     success = true
                     Cnvrg::API.parse_version(response)
+                    if response.to_hash[:status].to_i != 200
+                      Cnvrg::Logger.log_info("Got back bad status #{response.to_hash[:status]}")
+                    end
+                    if [503, 502, 429].include?(response.to_hash[:status].to_i)
+                      Cnvrg::Logger.log_info("Got back status #{response.to_hash[:status]}, will retry in #{5 * retries} seconds")
+                      success = false
+                      sleep(5 * retries)
+                      retries +=1
+                      next
+                    end
                   rescue => e
                     Cnvrg::Logger.log_error(e)
                     sleep(5)
@@ -95,11 +105,20 @@ module Cnvrg
                         response = conn.put "#{resource}", data.to_json if method.eql? 'PUT'
                         success = true
                         Cnvrg::API.parse_version(response)
-
+                        if response.to_hash[:status].to_i != 200
+                          Cnvrg::Logger.log_info("Got back bad status #{response.to_hash[:status]}")
+                        end
+                        if [503, 502, 429].include?(response.to_hash[:status].to_i)
+                          Cnvrg::Logger.log_info("Got back status #{response.to_hash[:status]}, will retry in #{5 * retries} seconds")
+                          success = false
+                          sleep(5 * retries)
+                          retries +=1
+                          next
+                        end
                     rescue => e
                       Cnvrg::Logger.log_error(e)
-                        sleep(5)
-                        retries +=1
+                      sleep(5)
+                      retries +=1
                     end
                 end
                 if !success

data/lib/cnvrg/cli.rb

@@ -3382,10 +3382,10 @@ module Cnvrg
                 if @project.is_git
                   output_dir = output_dir || @exp.output_dir
                   if output_dir.present?
-                    upload(false, false, true, ignore, true, true, output_dir, "Experiment", @exp.slug, true )
+                    upload(false, false, true, ignore, true, false, output_dir, "Experiment", @exp.slug, true )
                   end
                 else
-                  upload(false, false, true, ignore, true, true, nil, "Experiment", @exp.slug, true )
+                  upload(false, false, true, ignore, true, false, nil, "Experiment", @exp.slug, true )
                 end
               end
 
@@ -3652,6 +3652,7 @@ module Cnvrg
                 sleep 3
                 tries += 1
                 retry if tries <= 5
+                exit(1)
               end
             end
           end

data/lib/cnvrg/datafiles.rb

@@ -11,6 +11,7 @@ module Cnvrg
 
     LARGE_FILE=1024*1024*5
     MULTIPART_SPLIT=10000000
+    RETRIES = ENV['UPLOAD_FILE_RETRIES'].try(:to_i) || 10
 
     attr_reader :base_resource
 
@@ -256,6 +257,7 @@ module Cnvrg
     end
 
     def delete_file_chunk(commit_sha1, regex_list, chunk_size, offset)
+      retry_count = 0
       begin
         resp = Cnvrg::API.request(
             @base_resource + "delete_files_by_chunk",
@@ -268,13 +270,19 @@ module Cnvrg
             }
         )
         unless Cnvrg::CLI.is_response_success(resp, false)
-          Cnvrg::Logger.log_method(bind: binding)
           raise Exception.new("Got an error message from server, #{resp.try(:fetch, "message")}")
         end
         return resp["total_changes"]
       rescue => e
         Cnvrg::Logger.log_method(bind: binding)
         Cnvrg::Logger.log_error(e)
+
+        if retry_count < RETRIES
+          sleep(2**retry_count) # Exponential backoff
+          retry_count += 1
+          retry
+        end
+
         raise e
       end
     end

data/lib/cnvrg/downloader/client.rb

@@ -14,11 +14,13 @@ module Cnvrg
       end
 
       def extract_key_iv(sts_path)
-        count = 20
+        count = 0
         begin
           count += 1
           sts = open(sts_path, {ssl_verify_mode: 0}).read rescue nil
         rescue => e
+          backoff_time_seconds = backoff_time(count)
+          sleep backoff_time_seconds
           Cnvrg::Logger.log_error(e)
           retry if count <= 20
           raise StandardError.new("Cant access storage: #{e.message}")

data/lib/cnvrg/files.rb

@@ -733,7 +733,7 @@ module Cnvrg
       unless Cnvrg::CLI.is_response_success(res, false)
         raise SignalException.new("Cant download files from the server.")
       end
-      self.download_multpile_files_s3(res['result'], @project_home, postfix: postfix, progress: progress, threads: threads)
+      self.download_multiple_files_s3(res['result'], @project_home, postfix: postfix, progress: progress, threads: threads)
     end
 
 
@@ -750,7 +750,8 @@ module Cnvrg
       conflicted.each{|file| self.delete_conflict(file); progress.progress += 1 if progress.present?}
     end
 
-    def download_multpile_files_s3(files, project_home, postfix: '', progress: nil, threads: 15)
+    def download_multiple_files_s3(files, project_home, postfix: '', progress: nil, threads: 15)
+      cli = Cnvrg::CLI.new()
       begin
         props = {}
         client = props[:client]
@@ -774,11 +775,15 @@ module Cnvrg
           file_path = f["name"]
           if file_path.end_with? "/"
             # dir
-            if download_dir(file_path, file_path, project_home)
-              download_succ_count += 1
-            else
-              return Cnvrg::Result.new(false,"Could not create directory: #{file_path}")
-              raise Parallel::Kill
+            begin
+              if download_dir(file_path, file_path, project_home)
+                download_succ_count += 1
+              else
+                raise StandardError.new("Could not create directory #{file_path}.")
+              end
+            rescue => e
+              cli.log_message("Could not create directory #{file_path}. error: #{e.message}", Thor::Shell::Color::RED)
+              raise e
             end
           else
             file_path += postfix
@@ -793,25 +798,20 @@ module Cnvrg
               progress.progress += 1 if progress.present?
               download_succ_count += 1
             rescue => e
-              return Cnvrg::Result.new(false,"Could not create file: #{file_path}", e.message, e.backtrace)
-              raise Parallel::Kill
+              cli.log_message("Could not download file #{file_path}. error: #{e.message}", Thor::Shell::Color::RED)
+              raise e
             end
-
-
-
           end
         end
         if download_succ_count == files["keys"].size
           return Cnvrg::Result.new(true,"Done.\nDownloaded #{download_succ_count} files")
         end
       rescue => e
-        return Cnvrg::Result.new(false,"Could not download some files", e.message, e.backtrace)
+        cli.log_error(e)
+        raise e
       end
-
-
-
-
     end
+
     def download_file(absolute_path, relative_path, project_home, conflict=false)
         res = Cnvrg::API.request(@base_resource + "download_file", 'POST', {absolute_path: absolute_path, relative_path: relative_path})
         Cnvrg::CLI.is_response_success(res, false)

data/lib/cnvrg/helpers/agent.rb

@@ -75,9 +75,11 @@ class Cnvrg::Helpers::Agent
 
   def exec!
     log_internal("Command: #{@command} with slug: #{@slug} started!")
-    if should_run?
+    if @command.blank?
+      @exit_status = 0
+    elsif should_run?
       send_logs(status: Status::STARTED)
-      periodic_thread
+      periodic_thread_handle = periodic_thread
       execute_command
     else
       @exit_status = 127
@@ -86,6 +88,9 @@ class Cnvrg::Helpers::Agent
     finish_log += " after #{@real_execution_retries} retries" if @real_execution_retries > 0
     log_internal(finish_log)
     send_logs(exit_status: @exit_status, status: Status::FINISHED)
+    if periodic_thread_handle.present?
+      periodic_thread_handle.join
+    end
   end
 
   def get_logs_to_send

data/lib/cnvrg/helpers/executer.rb

@@ -120,17 +120,31 @@ class Cnvrg::Helpers::Executer
   end
 
   def init
-    resp = Cnvrg::API.request(activity_url, "PUT", {stats: executer_stats})
-    machine_activity = resp["machine_activity"]
-    Cnvrg::Logger.log_info("Got back machine activity #{machine_activity}")
-    if machine_activity.present? and @machine_activity != machine_activity
-      Cnvrg::Logger.log_info("Changing to machine activity #{machine_activity}")
-      machine_activity_yml = {slug: machine_activity}
-      File.open("/conf/.machine_activity.yml", "w+") {|f| f.write machine_activity_yml.to_yaml}
-      @machine_activity = machine_activity
+    retries = 0
+    success = false
+    puts("Agent started, connecting to #{Cnvrg::API.get_api}")
+    STDOUT.flush
+    while !success and retries < 100
+      begin
+        resp = Cnvrg::API.request(activity_url, "PUT", {stats: executer_stats})
+        machine_activity = resp["machine_activity"]
+        success = true
+        puts("Connected to server")
+        STDOUT.flush
+        Cnvrg::Logger.log_info("Got back machine activity #{machine_activity}")
+        if machine_activity.present? and @machine_activity != machine_activity
+          Cnvrg::Logger.log_info("Changing to machine activity #{machine_activity}")
+          machine_activity_yml = {slug: machine_activity}
+          File.open("/conf/.machine_activity.yml", "w+") {|f| f.write machine_activity_yml.to_yaml}
+          @machine_activity = machine_activity
+        end
+      rescue => e
+        Cnvrg::Logger.log_error(e)
+        Cnvrg::Logger.info("Sleeping for #{5 * retries}")
+        sleep(5 * retries)
+        retries +=1
+      end
     end
-  rescue => e
-    Cnvrg::Logger.log_error(e)
   end
 
   def polling_thread

data/lib/cnvrg/version.rb

@@ -1,3 +1,3 @@
 module Cnvrg
-  VERSION = '1.11.9'
+  VERSION = '1.11.15'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: cnvrg
 version: !ruby/object:Gem::Version
-  version: 1.11.9
+  version: 1.11.15
 platform: ruby
 authors:
 - Yochay Ettun
@@ -10,7 +10,7 @@ authors:
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2020-11-30 00:00:00.000000000 Z
+date: 2021-01-25 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler