cnvrg 1.9.9.9 → 1.10.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9c3589f491e89ed7e76b625e215b48cd98af828e2de0c3015435d452717740c3
-  data.tar.gz: a32b3b3dac6a0932a5dffe4a900a4f026b01ae524c5f60bc716d44553565eaa4
+  metadata.gz: 2c24bbf4a254e135ea5de9795e5394c0b480ddfdce161faeb60db440c398f129
+  data.tar.gz: 292d5f3fac1c85ed16d2fbb254c8d8296246f0672f885cfb8723cda155e92272
 SHA512:
-  metadata.gz: a9aa02bb0bcb739fe99f423d97bbec6536405588979f6faa5cda6054803419c5112b22a358f833e96981caa912e9e30d65db3cb4554b4d734dd27f59a65d8c45
-  data.tar.gz: 56bf19f2bdde12f8bb3a3f17efe185ecd5ca7edd7050e3fefd0bcb7fa237f800dbbbafb78974135e3e576396aa9d1685e44be19c3ca8491ad8ce90fa6c5fa545
+  metadata.gz: 44ffaa361515845b54a148d727385c224b29e539870abe2c7da88ab9357a9a69ae021ed1d3201bba92875895707c43aaa11c225918ac331c59b31086bf97b13e
+  data.tar.gz: fa2cec0111bf41e16efa30ae14e9c343fa2a29c8678660a2b7246ea3a1e3a1bb9a58ccf21955da849fc0362d9d1c756df7b8975ec497fdab5918f973d302aeeb

data/lib/cnvrg/api.rb

@@ -212,7 +212,6 @@ module Cnvrg
                return nil
             rescue SignalException
                 return false
-
             end
 
         end

data/lib/cnvrg/api_v2.rb

@@ -2,13 +2,76 @@ module Cnvrg
   class API_V2 < API
     ENDPOINT_VERSION = 'v2'
 
+    class CnvrgAPIError < StandardError; end
+
     def self.endpoint_uri
       api = get_api()
       return "#{api}/#{Cnvrg::API_V2::ENDPOINT_VERSION}"
     end
 
+    def self.request(resource, method = 'GET', data = {}, parse_request = true)
+      resource = URI::encode resource
+      n = Netrc.read
+      if n['cnvrg.io'].nil?
+        puts 'You\'re not logged in'
+        puts 'Please log in via `cnvrg login`'
+        return
+      end
+
+      _, pass = n[Cnvrg::Helpers.netrc_domain]
+
+      conn = Faraday.new endpoint_uri, :ssl => {:verify => !!Helpers.is_verify_ssl}
+      conn.headers['Auth-Token'] = pass
+      conn.headers['User-Agent'] = Cnvrg::API::USER_AGENT
+      conn.headers['Content-Type'] = "application/json"
+      conn.options.timeout = 420
+      conn.options.open_timeout = 180
+
+      20.times do
+        begin
+          response = send_request conn, resource, method, data
+          Cnvrg::API.parse_version response
+          is_response_success response
+          if parse_request
+            return JSON.parse(response.body)
+          else
+            return response.body
+          end
+        rescue CnvrgAPIError => e
+          raise e
+        rescue => e
+          Cnvrg::Logger.log_error e
+          sleep 0.5
+          retry
+        end
+      end
+    rescue CnvrgAPIError => e
+      Cnvrg::Logger.log_error e
+      raise e
+    rescue SignalException
+      return false
+    rescue => e
+      Cnvrg::Logger.log_error e
+      return nil
+    end
+
+    private
+
+    def self.send_request(conn, resource, method, data)
+      case method
+      when 'GET'
+        conn.get resource, data
+      when 'POST'
+        conn.post resource, data.to_json
+      when 'PUT'
+        conn.put resource, data.to_json
+      when 'DELETE'
+        conn.delete resource, data
+      end
+    end
+
     def self.is_response_success(response)
-      raise Exception.new("Bad status in response #{response.status}") if response.status != 200
+      raise CnvrgAPIError.new(JSON(response.body)['errors']) if response.status != 200
     end
   end
-end
+end

data/lib/cnvrg/cli.rb

@@ -49,6 +49,7 @@ require 'cnvrg/job_cli'
 require 'cnvrg/job_ssh'
 require 'cnvrg/connect_job_ssh'
 require 'cnvrg/api_v2'
+require 'rubygems/package'
 
 class Thor
   module Base
@@ -857,7 +858,7 @@ module Cnvrg
     method_option :read, :type => :boolean, :aliases => ["-r", "--read"], :default => false
     method_option :remote, :type => :boolean, :aliases => ["-h", "--remote"], :default => false
     method_option :soft, :type => :boolean, :aliases => ["-s", "--soft"], :default => false, :hide => true
-    def clone_data(dataset_url, only_tree=false, commit=nil, query=nil, read=false, remote=false, flatten: false, relative: false, soft: false)
+    def clone_data(dataset_url, only_tree=false, commit=nil, query=nil, read=false, remote=false, flatten: false, relative: false, soft: false, threads: 15)
       begin
         verify_logged_in(false)
         log_start(__method__, args, options)
@@ -913,7 +914,7 @@ module Cnvrg
 
           while files['keys'].length > 0
             Cnvrg::Logger.log_info("download multiple files, #{downloaded_files.size} files downloaded")
-            @files.download_multiple_files_s3(files, @dataset.local_path, progressbar: progressbar, read_only: read, flatten: flatten)
+            @files.download_multiple_files_s3(files, @dataset.local_path, progressbar: progressbar, read_only: read, flatten: flatten, threads: threads)
 
             downloaded_files += files['keys'].length
             files = @files.get_clone_chunk(commit: commit, latest_id: files['latest'])
@@ -1846,6 +1847,7 @@ module Cnvrg
     method_option :remote, :type => :boolean, :aliases => ["-r", "--r"], :default => false
     method_option :commit, :type => :string, :aliases => ["-c", "--c"], :default => nil
     method_option :soft, :type => :boolean, :aliases => ["-s", "--soft"], :default => false, :hide => true
+    method_option :threads, :type => :numeric, :aliases => ["--threads"], :default => 15
     def clone(project_url)
       begin
         verify_logged_in(false)
@@ -1856,7 +1858,7 @@ module Cnvrg
         owner = url_parts[project_index - 1]
         remote = options["remote"] || false
         soft = options["soft"] || false
-
+        threads = options[:threads] || Cnvrg::Helpers.parallel_threads
 
         response = Cnvrg::API.request("users/#{owner}/projects/#{slug}/get_project", 'GET')
         Cnvrg::CLI.is_response_success(response)
@@ -1907,9 +1909,10 @@ module Cnvrg
           idx = {commit: response["result"]["commit"], tree: response["result"]["tree"]}
           log_message("Downloading files", Thor::Shell::Color::BLUE)
           progressbar = @files.create_progressbar(files.size, "Clone Progress")
-          @files.download_files(files, commit_sha1, progress: progressbar)
+          @files.download_files(files, commit_sha1, progress: progressbar, threads: threads)
           progressbar.finish
           Project.verify_cnvrgignore_exist(project_name, remote)
+          Cnvrg::Logger.log_info("Generating idx")
           @project.set_idx(idx)
           log_message("Done")
           log_message("Downloaded #{files.size} files")
@@ -2274,13 +2277,16 @@ module Cnvrg
     method_option :job_type, :type => :string, :aliases => [ "--job_type"], :default => nil, :hide=>true
     method_option :suppress_exceptions, :type => :boolean, :aliases => ["--suppress-exceptions"], :default => true
     method_option :debug_mode, :type => :boolean, :aliases => ["--debug-mode"], :default => false
+    method_option :chunk_size, :type => :numeric, :aliases => ["--chunk"], :default => 1000
 
-    def upload(link = false, sync = false, direct = false, ignore_list = "", in_exp = false, force = false, output_dir = "output", job_type = nil, job_slug = nil, suppress_exceptions = true)
+    def upload(link = false, sync = false, direct = false, ignore_list = "", in_exp = false, force = false, output_dir = "output", job_type = nil, job_slug = nil, suppress_exceptions = true,chunk_size=1000)
       begin
         # we are passing "force" twice.. doesnt really make sense :\\
         verify_logged_in(true)
         log_start(__method__, args, options)
         @project = Project.new(get_project_home)
+        chunk_size = chunk_size ? chunk_size : options["chunk_size"]
+
 
         # Enable local/experiment exception logging
         suppress_exceptions = suppress_exceptions ? suppress_exceptions : options[:suppress_exceptions]
@@ -2396,7 +2402,7 @@ module Cnvrg
                                          :total => (to_upload.size + deleted.size),
                                          :autofinish => true)
 
-        buffered_errors = @files.upload_multiple_files(to_upload, commit_sha1, progress: progressbar, suppress_exceptions: suppress_exceptions)
+        buffered_errors = @files.upload_multiple_files(to_upload, commit_sha1, progress: progressbar, suppress_exceptions: suppress_exceptions, chunk_size: chunk_size)
         @files.delete_files_from_server(deleted, commit_sha1, suppress_exceptions: suppress_exceptions)
 
         progressbar.finish
@@ -2970,7 +2976,7 @@ module Cnvrg
     method_option :git_diff, :type => :boolean, :aliases => ["--git_diff"], :default => false
     method_option :suppress_exceptions, :type => :boolean, :aliases => ["--suppress-exceptions"], :default => true
     method_option :debug_mode, :type => :boolean, :aliases => ["--debug-mode"], :default => false
-    method_option :git_diff, :type => :boolean, :aliases => ["--git_diff"], :default => false
+    method_option :chunk_size, :type => :numeric, :aliases => ["--chunk"], :default => 1000
 
     def sync(direct = true)
       verify_logged_in(true) if direct
@@ -2995,7 +3001,8 @@ module Cnvrg
       end
       invoke :upload, [false, true,  direct, "",in_exp,options[:force], output_dir, job_type, job_slug ], :new_branch => options["new_branch"], :verbose => options["verbose"], :sync => true,
              :ignore => options[:ignore], :force => options[:force], :message => options[:message], :deploy => options["deploy"], :return_id => options["return_id"],
-             :files => options["files"], :output_dir => output_dir, :job_slug => job_slug, :job_type => job_type, :suppress_exceptions => options["suppress_exceptions"], :debug_mode => options['debug_mode'], :git_diff => options["git_diff"]
+             :files => options["files"], :output_dir => output_dir, :job_slug => job_slug, :job_type => job_type, :suppress_exceptions => options["suppress_exceptions"],
+             :debug_mode => options['debug_mode'], :git_diff => options["git_diff"], :chunk_size => options["chunk_size"]
 
     end
 
@@ -3039,6 +3046,7 @@ module Cnvrg
     method_option :notify_on_error, :type => :boolean, :aliases => ["-noe", "--notify_on_error"], :default => nil
     method_option :notify_on_success, :type => :boolean, :aliases => ["-nos", "--notify_on_success"], :default => nil
     method_option :emails, :type => :string, :aliases => ["-es", "--emails"], :default => "", :desc => "additional emails to notify on success / or error, comma separated"
+    method_option :wait, :type => :boolean, :aliases => ["-w", "--wait"], :default => false, :desc => "keep command session open until experiment finished to return exit status"
 
     def run(*cmd)
       verify_logged_in(true)
@@ -3073,6 +3081,7 @@ module Cnvrg
       email_notification_error = options["notify_on_error"]
       email_notification_success = options["notify_on_success"]
       emails = options["emails"]
+      wait = options["wait"]
 
       if !data.present? and data_query.present?
         log_message("Please provide data with data_query", Thor::Shell::Color::RED)
@@ -3118,14 +3127,14 @@ module Cnvrg
         invoke :exec_remote, [cmd], :sync_before => sync_before, :sync_after => sync_after, :title => title, :machine_type => instance_type,
                :schedule => schedule, :recurring => recurring, :log => log, :email_notification => email_notification, :upload_output => upload_output, :commit => commit,
                :image => image, :grid => grid, :data => data, :data_commit => data_commit, :ignore => ignore, :force => force, :sync_before_terminate => sync_before_terminate,
-                :max_time => max_time,
+               :max_time => max_time,
                :periodic_sync => periodic_sync, :dataset_only_tree=> dataset_only_tree,
                :output_dir=>output_dir, :data_query=>data_query, :git_commit =>git_commit, :git_branch=> git_branch,
                :restart_if_stuck =>restart_if_stuck, :local_folders => local_folders, :datasets => datasets, :prerun => prerun, :requirements => requirements,
-               :email_notification_error => email_notification_error, :email_notification_success => email_notification_success, :emails => emails
+               :email_notification_error => email_notification_error, :email_notification_success => email_notification_success, :emails => emails, :wait => wait
+
         return
       end
-
     end
 
     desc '', '', :hide => true
@@ -3288,7 +3297,7 @@ module Cnvrg
                       puts line
                     end
                     log << cur_log
-                    if log.size >= 1
+                    if log.size >= 10
                       @exp.upload_temp_log(log) unless log.empty?
                       log = []
                       elsif (start_time + 15.seconds) <= Time.now
@@ -3445,6 +3454,7 @@ module Cnvrg
     method_option :email_notification_error, :type => :boolean, :aliases => ["-noe", "--email_notification_error"], :default => true
     method_option :email_notification_success, :type => :boolean, :aliases => ["-nos", "--email_notification_success"], :default => true
     method_option :emails, :type => :string, :aliases => ["-es", "--emails"], :default => "", :desc => "additional emails to notify on success / or error"
+    method_option :wait, :type => :boolean, :aliases => ["-w", "--wait"], :default => false, :desc => "keep command session open until experiment finished to return exit status"
 
     def exec_remote(*cmd)
 
@@ -3506,7 +3516,7 @@ module Cnvrg
         options_hash.except!("schedule", "recurring", "machine_type", "image", "upload_output", "grid", "data", "data_commit", "title",
                              "local", "small", "medium", "large", "gpu", "gpuxl", "gpuxxl","max_time","dataset_only_tree",
                              "data_query", "git_commit","git_branch", "restart_if_stuck","local_folders","output_dir", "commit", "datasets",
-                             "requirements", "prerun", "email_notification_error", "email_notification_success", "emails")
+                             "requirements", "prerun", "email_notification_error", "email_notification_success", "emails", "wait")
         exec_options = options_hash.map {|x| "--#{x[0]}=#{x[1]}"}.flatten.join(" ")
         command = "#{exec_options} #{remote}  #{upload_output_option} #{cmd.flatten.join(" ")}"
         commit_to_run = options["commit"] || nil
@@ -3584,11 +3594,37 @@ module Cnvrg
           end
 
           log_message(str, Thor::Shell::Color::GREEN)
+          
+          exit_status = 0
+          
+          if options['wait']
+            while true
+              tries = 0
+              begin
+                result = Cnvrg::API_V2.request(
+                    "#{project.owner}/projects/#{project.slug}/experiments/#{res["result"]["exp_url"]}/status"
+                )
+                exit_status = result["exit_status"]
+                if exit_status
+                  puts "Experiment was exited with status: #{exit_status}"
+                  break
+                else
+                  system("clear") || system("cls")
+                  puts "#{Time.current}: waiting for experiment to finish"
+                  sleep 3
+                end
+              rescue => e
+                log_error(e)
+                log_message("Error occurred, retrying", Thor::Shell::Color::RED)
+                sleep 3
+                tries += 1
+                retry if tries <= 5
+              end
+            end
+          end
 
-          exit(0)
-          # end
+          exit(exit_status.to_i)
         end
-
       rescue => e
         log_message("Error occurred, Aborting", Thor::Shell::Color::RED)
         log_error(e)
@@ -3604,6 +3640,7 @@ module Cnvrg
         exit(1)
       end
     end
+
     desc 'deploy', 'Deploys model to production'
     method_option :small, :type => :boolean, :aliases => ["-s", "--small"], :default => false
     method_option :medium, :type => :boolean, :aliases => ["-m", "--medium"], :default => false

data/lib/cnvrg/data.rb

@@ -154,6 +154,7 @@ module Cnvrg
     method_option :relative, :type => :boolean, :aliases => ["-rel", "--relative"], :default => false
     method_option :flatten, :type => :boolean, :aliases => ["-f", "--flatten"], :default => false
     method_option :soft, :type => :boolean, :aliases => ["-s", "--soft"], :default => false, :hide => true
+    method_option :threads, :type => :numeric, :aliases => ["--threads"], :default => 15
     def clone(dataset_url)
       cli = Cnvrg::CLI.new()
       only_tree =options[:only_tree]
@@ -163,6 +164,7 @@ module Cnvrg
       remote = options[:remote]
       soft = options[:soft]
       flatten = options[:flatten]
+      threads = options[:threads]
       cli.clone_data(
           dataset_url,
           only_tree=only_tree,
@@ -172,7 +174,8 @@ module Cnvrg
           remote=remote,
           flatten: flatten,
           relative: options[:relative],
-          soft: soft
+          soft: soft,
+          threads: threads
       )
     end
 

data/lib/cnvrg/datafiles.rb

@@ -1284,11 +1284,11 @@ module Cnvrg
       end
     end
 
-    def download_multiple_files_s3(files, project_home, conflict: false, progressbar: nil, read_only:false, flatten: false)
+    def download_multiple_files_s3(files, project_home, conflict: false, progressbar: nil, read_only:false, flatten: false, threads: 15)
       begin
         refresh_storage_token
         parallel_options = {
-            in_threads: ParallelThreads,
+            in_threads: threads,
             isolation: true
         }
         Parallel.map(files["keys"], parallel_options) do |f|

data/lib/cnvrg/files.rb

@@ -86,7 +86,7 @@ module Cnvrg
       end
     end
 
-    def upload_multiple_files(files_list, commit_sha1, progress: nil, suppress_exceptions: false)
+    def upload_multiple_files(files_list, commit_sha1, progress: nil, suppress_exceptions: false, chunk_size: 100)
       #open files on the server.
       Cnvrg::Logger.log_info("Uploading project files")
       return if files_list.blank?
@@ -95,56 +95,62 @@ module Cnvrg
         return self.upload_files_old(files_list, commit_sha1, progress: progress)
       end
 
-      files_list = files_list.map{|x| [x, self.parse_file(x)] if self.parse_file(x)}.compact.to_h
-      resp = Cnvrg::API.request(@base_resource + "upload_files", 'POST', {files: files_list, commit: commit_sha1})
-      unless Cnvrg::CLI.is_response_success(resp, false)
-        raise SignalException.new("Cant upload files to the server.")
-      end
-      # resolve bucket
-      res = resp['result']
-      files = res['files']
-
-      #upload files
-      token_mutex = Mutex.new
+      blob_ids = []
       buffered_errors = {}
-      blob_ids = Parallel.map(files.keys, in_threads: ParallelThreads) do |file|
-
-        token_mutex.synchronize {
-          refresh_storage_token
-        }
-
-        begin
-          Cnvrg::Helpers.try_until_success{self.upload_single_file(files[file].merge(files_list[file]))}
-        rescue => e
-          Cnvrg::CLI.log_message("Failed to upload #{file}: #{e.message}", 'red') unless suppress_exceptions
-          Cnvrg::Logger.log_error(e)
-          Cnvrg::Logger.log_method(bind: binding)
-
-          buffered_errors[file] = "Failed to upload #{file}: #{e.message}" if suppress_exceptions
-
-          raise e unless suppress_exceptions
+      files_list.each_slice(chunk_size).each do |chunk_of_files|
+        Cnvrg::Logger.log_info("Upload chunk")
+        parsed_chunk_of_files = chunk_of_files.map{|x| [x, self.parse_file(x)] if self.parse_file(x)}.compact.to_h
+
+        resp = Cnvrg::API.request(@base_resource + "upload_files", 'POST', {
+            files: parsed_chunk_of_files,
+            commit: commit_sha1
+        })
+        unless Cnvrg::CLI.is_response_success(resp, false)
+          raise SignalException.new("Cant upload files to the server.")
         end
-        progress.progress += 1 if progress.present?
+        # resolve bucket
+        res = resp['result']
+        files = res['files']
+
+        #upload files
+        blob_id_chunk = Parallel.map(files.keys, in_threads: ParallelThreads) do |file|
+          begin
+            Cnvrg::Helpers.try_until_success{self.upload_single_file(files[file].merge(parsed_chunk_of_files[file]))}
+          rescue => e
+            Cnvrg::CLI.log_message("Failed to upload #{file}: #{e.message}", 'red') unless suppress_exceptions
+            Cnvrg::Logger.log_error(e)
+            Cnvrg::Logger.log_method(bind: binding)
+
+            buffered_errors[file] = "Failed to upload #{file}: #{e.message}" if suppress_exceptions
+
+            raise e unless suppress_exceptions
+          end
+          progress.progress += 1 if progress.present?
 
-        unless buffered_errors.key?(file)
-          files[file]["bv_id"]
-        else
-          nil
+          unless buffered_errors.key?(file)
+            files[file]["bv_id"]
+          else
+            nil
+          end
         end
-      end
 
-      # remove nil files (failed files) from blob_ids
-      blob_ids.compact!
 
-      #save files on the server.
-      resp = Cnvrg::API.request(@base_resource + "upload_files_save", 'POST', {blob_ids: blob_ids, commit: commit_sha1})
-      unless Cnvrg::CLI.is_response_success(resp, false)
-        raise SignalException.new("Cant save uploaded files to the server.")
-      end
+      blob_ids.concat blob_id_chunk
+    end
+
+    # remove nil files (failed files) from blob_ids
+    blob_ids.compact!
 
-      return buffered_errors
+    #save files on the server.
+    resp = Cnvrg::API.request(@base_resource + "upload_files_save", 'POST', {blob_ids: blob_ids, commit: commit_sha1})
+    unless Cnvrg::CLI.is_response_success(resp, false)
+      raise SignalException.new("Cant save uploaded files to the server.")
     end
 
+    return buffered_errors
+  end
+
+
 
     def delete_files_from_server_old(files, commit_sha1)
       #files are absolute path here.
@@ -717,7 +723,7 @@ module Cnvrg
       @progressbar
     end
 
-    def download_files(files, commit, postfix: '', progress: nil)
+    def download_files(files, commit, postfix: '', progress: nil, threads: 15)
       return if files.blank?
       if Cnvrg::Helpers.server_version < 1
         Cnvrg::Logger.log_info("Download files from older server.")
@@ -727,7 +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)
+      self.download_multpile_files_s3(res['result'], @project_home, postfix: postfix, progress: progress, threads: threads)
     end
 
 
@@ -744,7 +750,7 @@ 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)
+    def download_multpile_files_s3(files, project_home, postfix: '', progress: nil, threads: 15)
       begin
         props = {}
         client = props[:client]
@@ -753,7 +759,7 @@ module Cnvrg
         bucket = props[:bucket]
         download_succ_count = 0
         parallel_options = {
-            in_threads: Cnvrg::Helpers.parallel_threads,
+            in_threads: threads,
             isolation: true
         }
 

data/lib/cnvrg/helpers.rb

@@ -2,15 +2,8 @@ module Cnvrg
   module Helpers
 
     extend self
-    def parallel_threads()
-      threads = ENV["CNVRG_PARALLEL_THREADS"].to_i
-
-      
-      ### if empty, default will be 15 threads
-      threads = threads > 0 ? threads : 15
-
-      ### set max threads to be 100k
-      [threads, 100000].min
+    def parallel_threads
+      15
     end
 
     def self.parallel_options

data/lib/cnvrg/version.rb

@@ -1,3 +1,3 @@
 module Cnvrg
-  VERSION = '1.9.9.9'
+  VERSION = '1.10.4'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: cnvrg
 version: !ruby/object:Gem::Version
-  version: 1.9.9.9
+  version: 1.10.4
 platform: ruby
 authors:
 - Yochay Ettun
@@ -10,7 +10,7 @@ authors:
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2020-07-20 00:00:00.000000000 Z
+date: 2020-08-04 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler