llm_logs 0.7.1 → 0.8.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 31867809a66b41a89465171c11b5706ef364f9276fbd145f9545cd7b5c27768f
-  data.tar.gz: 58dc6a88ba5b337420a380fee8525b0a84ff8e32612cc05c56fd1899ca91dbbc
+  metadata.gz: 5bc6872f875f8ceddec93621ab746e249266100a8bd1172645fc0d3f24bdf5b7
+  data.tar.gz: 842fdce69e1c3773e65adc38144738ae0dd2e7eb579d5826958fd42858fb6488
 SHA512:
-  metadata.gz: 83d8bca6f9869ef0e40e9862e04228e4a3830b9d5cd17d29335aad9422c5b401fe70e655a5e58cb0f0cdcdfe40e5de6487a1b144d6f37b78000feba19ed4039a
-  data.tar.gz: 94ab95a5f5213a5ab7dbd3294e887db24191bcb16d080928e394b3e51d53f00e7ae5a25a8a572401ccbd283c33a137d242c71dcbc174c33f64754a28cdfc9c6f
+  metadata.gz: 8e0eeda8eb30321b52abd6b0646f4f47b6a8b557265319372b3fd72cfa8f651a02ad95855952c15c4e2f312623d2b4c751fb50296467440f75bfa676cccccbe3
+  data.tar.gz: b403981ee9cd7313581d52f54d236bbf32341bb707e1f37f0c6b78dae74504a99171ed7839922ebaa9abd8589498a23c66c12aa9b0cc7bb2a5c95d0435e1b705

data/README.md

@@ -71,6 +71,12 @@ Each span records:
 
 Tool calls are captured as child spans with tool name and arguments.
 
+Each provider round resends the whole conversation, so a round that continues the trace's previous round stores only the messages it added. Its metadata names the round it continues (`input_base_span_id`) and how many of that round's messages it carried over (`input_offset`). `span.full_input` returns the full list. A round that doesn't start with the previous round's messages, such as a second chat in the same trace, stores its full input.
+
+### Retention
+
+`LlmLogs::PurgeJob` deletes traces started more than `retention_days` ago, together with their spans. Schedule it from the host app, for example once a day. With `retention_days = nil` it keeps everything.
+
 ## Prompt Management
 
 Create prompts with [Mustache](https://mustache.github.io/) templates and auto-versioning.
@@ -323,7 +329,7 @@ Browse traces and manage prompts at `/llm_logs`.
 LlmLogs.setup do |config|
   config.enabled = true                                      # master switch for logging
   config.auto_instrument = true                              # subscribe to ruby_llm notifications (needs RubyLLM.config.instrumenter = ActiveSupport::Notifications)
-  config.retention_days = 30                                 # for future cleanup job
+  config.retention_days = 30                                 # LlmLogs::PurgeJob deletes older traces; nil keeps them
   config.prompts_source_path = Rails.root.join("db/data/prompts")
   config.prompt_subfolders = %w[skills fragments templates]
   config.batch_enabled = true                                # enable the batch API integration

data/app/controllers/llm_logs/prompts_controller.rb

@@ -6,9 +6,10 @@ module LlmLogs
       tag = params[:tag].is_a?(String) ? params[:tag].presence : nil
       @sort      = SORT_COLUMNS.key?(params[:sort]) ? params[:sort] : "name"
       @direction = params[:direction] == "desc" ? "desc" : "asc"
-      scope = Prompt.order(SORT_COLUMNS.fetch(@sort) => @direction.to_sym).includes(:versions)
+      scope = Prompt.order(SORT_COLUMNS.fetch(@sort) => @direction.to_sym)
       scope = scope.with_tag(tag) if tag
       @prompts    = scope.page(params[:page]).per(LlmLogs.page_size)
+      @latest_versions = PromptVersion.latest_for(@prompts.map(&:id)).index_by(&:prompt_id)
       @active_tag = tag
       @all_tags   = Prompt.pluck(:tags).flatten.compact.uniq.sort
     end

data/app/controllers/llm_logs/spans_controller.rb

@@ -3,6 +3,7 @@ module LlmLogs
     def show
       @trace = Trace.find(params[:trace_id])
       @span = @trace.spans.find(params[:id])
+      @input = @span.full_input
     end
   end
 end

data/app/controllers/llm_logs/traces_controller.rb

@@ -12,8 +12,10 @@ module LlmLogs
 
     def show
       @trace = Trace.includes(prompt_version: :prompt).find(params[:id])
-      @root_spans = @trace.root_spans
-      @models = @trace.spans.where(span_type: "llm").distinct.pluck(:model).compact
+      @spans = @trace.spans.order(:started_at).to_a
+      @child_spans = @spans.group_by(&:parent_span_id)
+      @root_spans = @child_spans.fetch(nil, [])
+      @models = @spans.select { |span| span.span_type == "llm" }.filter_map(&:model).uniq
     end
 
   end

data/app/jobs/llm_logs/purge_job.rb

@@ -0,0 +1,21 @@
+module LlmLogs
+  # Deletes traces (and their spans) started more than LlmLogs.retention_days ago. Schedule it
+  # from the host app, e.g. daily; a nil retention_days keeps everything.
+  class PurgeJob < ::ActiveJob::Base
+    queue_as :default
+
+    BATCH_SIZE = 500
+
+    def perform
+      days = LlmLogs.retention_days
+      return if days.nil?
+
+      LlmLogs::Trace.where(started_at: ...days.days.ago).in_batches(of: BATCH_SIZE) do |traces|
+        trace_ids = traces.ids
+        LlmLogs::BatchRequest.where(trace_id: trace_ids).update_all(trace_id: nil)
+        LlmLogs::Span.where(trace_id: trace_ids).delete_all
+        LlmLogs::Trace.where(id: trace_ids).delete_all
+      end
+    end
+  end
+end

data/app/models/llm_logs/prompt_version.rb

@@ -9,6 +9,14 @@ module LlmLogs
     validates :messages, presence: true
     validate :reasoning_effort_is_supported
 
+    # The newest version of each given prompt, without its messages (for listings).
+    def self.latest_for(prompt_ids)
+      newest = unscoped.where(prompt_id: prompt_ids).group(:prompt_id).select(:prompt_id, "MAX(version_number) AS version_number")
+      select(:id, :prompt_id, :version_number, :model, :model_params)
+        .joins("INNER JOIN (#{newest.to_sql}) newest ON newest.prompt_id = #{table_name}.prompt_id " \
+               "AND newest.version_number = #{table_name}.version_number")
+    end
+
     def reasoning_effort
       return nil unless model_params.is_a?(Hash)
 

data/app/models/llm_logs/span.rb

@@ -9,6 +9,31 @@ module LlmLogs
     validates :status, presence: true, inclusion: { in: %w[ok error] }
     validates :started_at, presence: true
 
+    # An llm round that continues the trace's previous round stores only the messages it
+    # added (input) plus how many it carried over from that round (input_offset).
+    def input_base_span_id
+      metadata&.dig("input_base_span_id")
+    end
+
+    def input_offset
+      metadata&.dig("input_offset")
+    end
+
+    # The messages this round sent, rebuilt from the chain of rounds it continues. Spans
+    # recorded before llm_logs 0.8 have no base and already hold their full input.
+    def full_input
+      return input unless input_base_span_id
+
+      rounds = trace.spans.where(span_type: "llm").where(id: ...id).select(:id, :input, :metadata).index_by(&:id)
+      chain = [self]
+      while (base = rounds[chain.first.input_base_span_id])
+        chain.unshift(base)
+      end
+      chain.drop(1).reduce(Array(chain.first.input)) do |messages, span|
+        messages.first(span.input_offset) + Array(span.input)
+      end
+    end
+
     def finish
       update!(
         completed_at: Time.current,

data/app/models/llm_logs/trace.rb

@@ -38,10 +38,8 @@ module LlmLogs
     private
 
     def rollup_stats!
-      self.total_input_tokens = spans.sum(:input_tokens)
-      self.total_output_tokens = spans.sum(:output_tokens)
-      self.total_cached_tokens = spans.sum(:cached_tokens)
-      self.total_cost = spans.sum(:cost)
+      totals = spans.pick(*%w[input_tokens output_tokens cached_tokens cost].map { |column| Arel.sql("COALESCE(SUM(#{column}), 0)") })
+      self.total_input_tokens, self.total_output_tokens, self.total_cached_tokens, self.total_cost = totals
     end
   end
 end

data/app/views/llm_logs/prompts/index.html.erb

@@ -27,7 +27,7 @@
     </thead>
     <tbody class="divide-y divide-gray-200">
       <% @prompts.each do |prompt| %>
-        <% latest = prompt.versions.max_by(&:version_number) %>
+        <% latest = @latest_versions[prompt.id] %>
         <tr class="hover:bg-gray-50">
           <td class="px-4 py-3 text-sm">
             <%= link_to prompt.name, prompt_path(prompt), class: "text-indigo-600 hover:text-indigo-900 font-medium" %>

data/app/views/llm_logs/spans/show.html.erb

@@ -45,21 +45,21 @@
   </div>
 </div>
 
-<% if @span.input.present? %>
+<% if @input.present? %>
   <div class="bg-white rounded-lg shadow-sm ring-1 ring-gray-900/5 mb-6">
     <div class="px-4 py-3 border-b border-gray-200">
-      <h2 class="text-sm font-medium text-gray-900">Input<%= @span.input.is_a?(Array) ? " Messages" : "" %></h2>
+      <h2 class="text-sm font-medium text-gray-900">Input<%= @input.is_a?(Array) ? " Messages" : "" %></h2>
     </div>
     <div class="p-4 space-y-3">
-      <% if @span.input.is_a?(Array) %>
-        <% @span.input.each do |msg| %>
+      <% if @input.is_a?(Array) %>
+        <% @input.each do |msg| %>
           <div class="rounded-lg p-3 <%= msg['role'] == 'user' ? 'bg-blue-50' : 'bg-gray-50' %>">
             <div class="text-xs font-medium text-gray-500 uppercase mb-1"><%= msg['role'] %></div>
             <pre class="text-sm whitespace-pre-wrap"><%= msg['content'] %></pre>
           </div>
         <% end %>
       <% else %>
-        <pre class="text-sm bg-gray-50 rounded-lg p-3 whitespace-pre-wrap"><%= pretty_json(@span.input) %></pre>
+        <pre class="text-sm bg-gray-50 rounded-lg p-3 whitespace-pre-wrap"><%= pretty_json(@input) %></pre>
       <% end %>
     </div>
   </div>

data/app/views/llm_logs/traces/_span_node.html.erb

@@ -28,9 +28,15 @@
     </summary>
 
     <div class="mt-2 ml-4 space-y-3">
-      <% if span.input.present? %>
+      <% if span.input.present? || span.input_base_span_id %>
         <div>
           <h4 class="text-xs font-medium text-gray-500 uppercase mb-1">Input</h4>
+          <% if span.input_base_span_id %>
+            <p class="text-xs text-gray-500 mb-1">
+              New messages only; <%= pluralize(span.input_offset, "earlier message") %> from the previous round not repeated.
+              <%= link_to "Full input", trace_span_path(trace, span), class: "text-indigo-600 hover:text-indigo-900" %>
+            </p>
+          <% end %>
           <pre class="text-xs bg-gray-50 rounded p-3 max-h-64 overflow-y-auto border"><%= pretty_json(span.input) %></pre>
         </div>
       <% end %>
@@ -60,7 +66,7 @@
     </div>
   </details>
 
-  <% span.child_spans.order(:started_at).each do |child| %>
-    <%= render partial: "llm_logs/traces/span_node", locals: { span: child, depth: depth + 1, trace: trace } %>
+  <% child_spans.fetch(span.id, []).each do |child| %>
+    <%= render partial: "llm_logs/traces/span_node", locals: { span: child, depth: depth + 1, trace: trace, child_spans: child_spans } %>
   <% end %>
 </div>

data/app/views/llm_logs/traces/show.html.erb

@@ -32,7 +32,7 @@
   </div>
   <div class="bg-white rounded-lg p-4 shadow-sm ring-1 ring-gray-900/5">
     <dt class="text-xs font-medium text-gray-500 uppercase">Spans</dt>
-    <dd class="text-2xl font-semibold text-gray-900 mt-1"><%= @trace.spans.count %></dd>
+    <dd class="text-2xl font-semibold text-gray-900 mt-1"><%= @spans.size %></dd>
   </div>
   <div class="bg-white rounded-lg p-4 shadow-sm ring-1 ring-gray-900/5">
     <dt class="text-xs font-medium text-gray-500 uppercase">Input Tokens</dt>
@@ -76,7 +76,7 @@
   <div class="p-4">
     <% if @root_spans.any? %>
       <% @root_spans.each do |span| %>
-        <%= render partial: "llm_logs/traces/span_node", locals: { span: span, depth: 0, trace: @trace } %>
+        <%= render partial: "llm_logs/traces/span_node", locals: { span: span, depth: 0, trace: @trace, child_spans: @child_spans } %>
       <% end %>
     <% else %>
       <p class="text-sm text-gray-500">No spans recorded.</p>

data/lib/llm_logs/instrumentation/ruby_llm_chat.rb

@@ -105,14 +105,20 @@ module LlmLogs
 
             chat = payload[:chat]
             payload[USAGE_START_KEY] = chat.usage_entries.length if chat.respond_to?(:usage_entries)
-            payload[SPAN_KEY] = LlmLogs::Tracer.start_span(
+            messages = Serialize.messages(payload[:input_messages])
+            base = continued_round(messages)
+            metadata = request_metadata(payload)
+            metadata.merge!("input_base_span_id" => base[:span_id], "input_offset" => base[:messages].size) if base
+            span = LlmLogs::Tracer.start_span(
               name: "chat.complete",
               span_type: "llm",
               model: payload[:model],
               provider: payload[:provider],
-              input: Serialize.messages(payload[:input_messages]),
-              metadata: request_metadata(payload)
+              input: base ? messages.drop(base[:messages].size) : messages,
+              metadata: metadata
             )
+            LlmLogs::Tracer.last_llm_round = {trace_id: span.trace_id, span_id: span.id, messages: messages}
+            payload[SPAN_KEY] = span
           end
         end
 
@@ -137,6 +143,21 @@ module LlmLogs
 
         private
 
+        # Every round of a chat resends the whole conversation. When this round's messages
+        # start with the trace's previous round's, the span stores only the new ones and
+        # points at that round; Span#full_input puts them back together. Anything else
+        # (another chat in the same trace, a rewritten history) stores the full input.
+        def continued_round(messages)
+          base = LlmLogs::Tracer.last_llm_round
+          trace = LlmLogs::Tracer.current_trace
+          return nil unless base && trace && base[:trace_id] == trace.id
+
+          prefix = base[:messages]
+          return nil if prefix.empty? || prefix.size > messages.size
+
+          messages.first(prefix.size) == prefix ? base : nil
+        end
+
         def request_metadata(payload)
           {
             "streaming" => payload[:streaming],

data/lib/llm_logs/tracer.rb

@@ -17,6 +17,16 @@ module LlmLogs
       Fiber[:llm_logs_span] = span
     end
 
+    # The current trace's latest llm round, {trace_id:, span_id:, messages:}, kept so the next
+    # round can store only the messages it adds (see Instrumentation::RubyLlmChat).
+    def self.last_llm_round
+      Fiber[:llm_logs_last_llm_round]
+    end
+
+    def self.last_llm_round=(round)
+      Fiber[:llm_logs_last_llm_round] = round
+    end
+
     def self.start_trace(name, metadata: {})
       trace = LlmLogs::Trace.create!(
         name: name,
@@ -27,8 +37,10 @@ module LlmLogs
 
       previous_trace = Fiber[:llm_logs_trace]
       previous_span = Fiber[:llm_logs_span]
+      previous_round = Fiber[:llm_logs_last_llm_round]
       Fiber[:llm_logs_trace] = trace
       Fiber[:llm_logs_span] = nil
+      Fiber[:llm_logs_last_llm_round] = nil
 
       begin
         yield trace
@@ -39,6 +51,7 @@ module LlmLogs
         trace.complete! if trace.status == "running"
         Fiber[:llm_logs_trace] = previous_trace
         Fiber[:llm_logs_span] = previous_span
+        Fiber[:llm_logs_last_llm_round] = previous_round
       end
     end
 

data/lib/llm_logs/version.rb

@@ -1,3 +1,3 @@
 module LlmLogs
-  VERSION = "0.7.1"
+  VERSION = "0.8.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm_logs
 version: !ruby/object:Gem::Version
-  version: 0.7.1
+  version: 0.8.0
 platform: ruby
 authors:
 - Anton
@@ -189,6 +189,7 @@ files:
 - app/helpers/llm_logs/prompts_helper.rb
 - app/jobs/llm_logs/batch/flush_job.rb
 - app/jobs/llm_logs/batch/poll_job.rb
+- app/jobs/llm_logs/purge_job.rb
 - app/models/llm_logs/application_record.rb
 - app/models/llm_logs/batch.rb
 - app/models/llm_logs/batch/adapters/bedrock.rb