Class: ImageFullAnalysisWorker

Inherits:
Object
  • Object
show all
Includes:
Sidekiq::Job, Workers::StatusBroadcastable
Defined in:
app/workers/image_full_analysis_worker.rb

Overview

Background worker for running the complete AI analysis pipeline on an image.

== Pipeline (Gemini)

  1. pHash Fingerprint - Perceptual hash for duplicate detection (local)
  2. Gemini Embedding 2 - Native multimodal embedding (image + metadata text)
  3. Gemini Flash Vision - Describes image content (independent, for CRM/metadata)

The embedding step no longer depends on vision analysis — Gemini Embedding 2
natively understands images without a text intermediary.

Examples:

Queue full analysis

ImageFullAnalysisWorker.perform_async(123)

Constant Summary collapse

JOB_TRACKING_KEY =

Key used for job tracking.

"image_full_analysis_jid"
JOB_TRACKING_TTL =

Job tracking ttl.

2.hours.to_i
MODEL =

Model (GA) — sourced from the canonical registry.

AiModelConstants.id(:unified_embedding)
DIMENSIONS =

Dimensions.

1536

Instance Attribute Summary

Attributes included from Workers::StatusBroadcastable

#broadcast_status_updates

Class Method Summary collapse

Instance Method Summary collapse

Methods included from Workers::StatusBroadcastable::Overrides

#at, #store, #total

Class Method Details

.find_running_jid(image_id) ⇒ Object

Find running jid.

Parameters:

  • image_id (Integer) —

    the image id

Returns:

  • (Object) —

    the result



50
51
52
53
54
55
 
# File 'app/workers/image_full_analysis_worker.rb', line 50

def self.find_running_jid(image_id)
  jid = Sidekiq.redis { |r| r.get("#{JOB_TRACKING_KEY}:#{image_id}") }
  return nil unless jid

  Sidekiq::Status.status(jid).in?(%i[queued working]) ? jid : nil
end

.lock_args(args) ⇒ Array

Lock args.

Parameters:

  • args (Array) —

    the args to lock

Returns:

  • (Array) —

    the resulting array of values



28
29
30
 
# File 'app/workers/image_full_analysis_worker.rb', line 28

def self.lock_args(args)
  [args[0]]
end

.track_jid(image_id, jid) ⇒ void

This method returns an undefined value.

Track jid.

Parameters:

  • image_id (Integer) —

    the image id

  • jid (String) —

    the jid to track



42
43
44
 
# File 'app/workers/image_full_analysis_worker.rb', line 42

def self.track_jid(image_id, jid)
  Sidekiq.redis { |r| r.set("#{JOB_TRACKING_KEY}:#{image_id}", jid, ex: JOB_TRACKING_TTL) }
end

Instance Method Details

#perform(image_id, options = {}) ⇒ Object

Parameters:

  • image_id (Integer)
  • options (Hash) (defaults to: {}) —

    job options

Options Hash (options):

  • force (Boolean) —

    force re-analysis even if results are up to date

  • redirect_to (String) —

    path to redirect to when the job completes



70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
 
# File 'app/workers/image_full_analysis_worker.rb', line 70

def perform(image_id, options = {})
  @force = options[:force].to_b

  total 4
  at 0, 'Starting AI analysis...'

  image = Image.find_by(id: image_id)
  unless image
    store error_message: "Image #{image_id} not found"
    return log_info("Image #{image_id} not found")
  end

  redirect_to_path = options[:redirect_to]
  redirect_to_path ||= "/en-US/images/#{image.slug}/tab_embeddings?target_id=ai_embeddings"
  store redirect_to: redirect_to_path

  if image.inactive?
    store error_message: "Image #{image_id} is inactive"
    return log_info("Image #{image_id} inactive")
  end

  if image.asset&.dig('file_type') == 'non-image'
    store error_message: 'Non-image file type'
    return log_info("Image #{image_id} is non-image")
  end

  analysis = analyze_needs(image)
  log_info "Image #{image_id}: #{analysis}"

  unless analysis[:needs_anything]
    at 4, 'Already complete'
    store info_message: 'All up to date'
    return log_info("Image #{image_id}: Up to date")
  end

  # Step 1: pHash
  at 1, 'Step 1/3: pHash...'
  if analysis[:needs_phash]
    begin
      ImageFingerprintWorker.new.perform(image_id, { force: @force })
      log_info "Image #{image_id}: pHash complete"
    rescue StandardError => e
      log_error "Image #{image_id}: pHash failed: #{e.message}"
    end
  end

  # Step 2: Multimodal embedding (Gemini Embedding 2 — image + metadata text)
  at 2, 'Step 2/3: Multimodal embedding...'
  if analysis[:needs_embedding]
    begin
      generate_multimodal_embedding(image)
      log_info "Image #{image_id}: Multimodal embedding complete"
    rescue Embedding::Gemini::PermanentError => e
      # Broken/invalid source image (404 URL, corrupt file). Retrying can't fix
      # it, so QUARANTINE the asset (invalidate! excludes it from fetches +
      # embedding and starts the 30-day purge clock) and continue — do NOT
      # re-raise. Transient errors (rate limit, timeout, connection, 5xx) are
      # deliberately NOT rescued here: they propagate to the outer rescue, which
      # re-raises so Sidekiq RETRIES the embed instead of silently dropping it.
      log_error "Image #{image_id}: Embedding permanently failed (quarantining): #{e.message}"
      image.invalidate!(reason: "embedding: #{e.message.truncate(120)}")
    end
  end

  # Step 3: Vision description (Gemini Flash — independent, for CRM/metadata)
  at 3, 'Step 3/3: Vision analysis...'
  if analysis[:needs_vision]
    begin
      result = ImageAnalysis::VisionAnalyzer.call(image, force: @force)
      if result.success?
        log_info "Image #{image_id}: Vision analysis complete"
      else
        log_error "Image #{image_id}: Vision failed: #{result.error}"
      end
    rescue StandardError => e
      log_error "Image #{image_id}: Vision failed: #{e.message}"
    end
  end

  at 4, 'Complete!'
  store info_message: 'AI analysis complete'
  log_info "Image #{image_id}: Complete"
rescue Embedding::Gemini::RateLimitError, Faraday::TimeoutError, Faraday::ConnectionFailed => e
  # Transient (rate limit / network). Let Sidekiq retry with backoff; don't
  # spam error reporting for expected throttling.
  log_error "Image #{image_id}: transient embedding error, will retry: #{e.message}"
  raise
rescue StandardError => e
  store error_message: "Error: #{e.message}"
  log_error "Error for Image #{image_id}: #{e.message}"
  ErrorReporting.error(e)
  raise
end