Merge branch 'master' into browserstep-refactor

.github/workflows/test-only.yml

@@ -58,9 +58,9 @@ jobs:
           # restock detection via playwright - added name=changedet here so that playwright/browserless can connect to it
           docker run --rm --name "changedet" -e "FLASK_SERVER_NAME=changedet" -e "PLAYWRIGHT_DRIVER_URL=ws://browserless:3000" --network changedet-network test-changedetectionio  bash -c 'cd changedetectionio;pytest --live-server-port=5004 --live-server-host=0.0.0.0 tests/restock/test_restock.py'
 
-      - name: Test with puppeteer fetcher
+      - name: Test with puppeteer fetcher and disk cache
         run: |
-          docker run --rm -e "USE_EXPERIMENTAL_PUPPETEER_FETCH=yes" -e "PLAYWRIGHT_DRIVER_URL=ws://browserless:3000" --network changedet-network test-changedetectionio  bash -c 'cd changedetectionio;pytest tests/fetchers/test_content.py && pytest tests/test_errorhandling.py && pytest tests/visualselector/test_fetch_data.py'
+          docker run --rm -e "PUPPETEER_DISK_CACHE=/tmp/data/" -e "USE_EXPERIMENTAL_PUPPETEER_FETCH=yes" -e "PLAYWRIGHT_DRIVER_URL=ws://browserless:3000" --network changedet-network test-changedetectionio  bash -c 'cd changedetectionio;pytest tests/fetchers/test_content.py && pytest tests/test_errorhandling.py && pytest tests/visualselector/test_fetch_data.py'
           # Browserless would have had -e "FUNCTION_BUILT_INS=[\"fs\",\"crypto\"]" added above
 
       - name: Test proxy interaction

changedetectionio/blueprint/browser_steps/__init__.py

@@ -27,58 +27,103 @@ import os
 import logging
 from changedetectionio.store import ChangeDetectionStore
 from changedetectionio import login_optionally_required
-browsersteps_live_ui_o = {}
-browsersteps_playwright_browser_interface = None
-browsersteps_playwright_browser_interface_browser = None
-browsersteps_playwright_browser_interface_context = None
-browsersteps_playwright_browser_interface_end_time = None
-browsersteps_playwright_browser_interface_start_time = None
 
-def cleanup_playwright_session():
+browsersteps_sessions = {}
+io_interface_context = None
 
-    global browsersteps_live_ui_o
-    global browsersteps_playwright_browser_interface
-    global browsersteps_playwright_browser_interface_browser
-    global browsersteps_playwright_browser_interface_context
-    global browsersteps_playwright_browser_interface_end_time
-    global browsersteps_playwright_browser_interface_start_time
-
-    browsersteps_live_ui_o = {}
-    browsersteps_playwright_browser_interface = None
-    browsersteps_playwright_browser_interface_browser = None
-    browsersteps_playwright_browser_interface_end_time = None
-    browsersteps_playwright_browser_interface_start_time = None
-
-    print("Cleaning up old playwright session because time was up, calling .goodbye()")
-    try:
-        browsersteps_playwright_browser_interface_context.goodbye()
-    except Exception as e:
-        print ("Got exception in shutdown, probably OK")
-        print (str(e))
-
-    browsersteps_playwright_browser_interface_context = None
-
-    print ("Cleaning up old playwright session because time was up - done")
 
 def construct_blueprint(datastore: ChangeDetectionStore):
-
     browser_steps_blueprint = Blueprint('browser_steps', __name__, template_folder="templates")
 
+    def start_browsersteps_session(watch_uuid):
+        from . import nonContext
+        from . import browser_steps
+        import time
+        global browsersteps_sessions
+        global io_interface_context
+
+
+        # We keep the playwright session open for many minutes
+        seconds_keepalive = int(os.getenv('BROWSERSTEPS_MINUTES_KEEPALIVE', 10)) * 60
+
+        browsersteps_start_session = {'start_time': time.time()}
+
+        # You can only have one of these running
+        # This should be very fine to leave running for the life of the application
+        # @idea - Make it global so the pool of watch fetchers can use it also
+        if not io_interface_context:
+            io_interface_context = nonContext.c_sync_playwright()
+            # Start the Playwright context, which is actually a nodejs sub-process and communicates over STDIN/STDOUT pipes
+            io_interface_context = io_interface_context.start()
+
+
+        # keep it alive for 10 seconds more than we advertise, sometimes it helps to keep it shutting down cleanly
+        keepalive = "&timeout={}".format(((seconds_keepalive + 3) * 1000))
+        try:
+            browsersteps_start_session['browser'] = io_interface_context.chromium.connect_over_cdp(
+                os.getenv('PLAYWRIGHT_DRIVER_URL', '') + keepalive)
+        except Exception as e:
+            if 'ECONNREFUSED' in str(e):
+                return make_response('Unable to start the Playwright Browser session, is it running?', 401)
+            else:
+                return make_response(str(e), 401)
+
+        proxy_id = datastore.get_preferred_proxy_for_watch(uuid=watch_uuid)
+        proxy = None
+        if proxy_id:
+            proxy_url = datastore.proxy_list.get(proxy_id).get('url')
+            if proxy_url:
+
+                # Playwright needs separate username and password values
+                from urllib.parse import urlparse
+                parsed = urlparse(proxy_url)
+                proxy = {'server': proxy_url}
+
+                if parsed.username:
+                    proxy['username'] = parsed.username
+
+                if parsed.password:
+                    proxy['password'] = parsed.password
+
+                print("Browser Steps: UUID {} selected proxy {}".format(watch_uuid, proxy_url))
+
+        # Tell Playwright to connect to Chrome and setup a new session via our stepper interface
+        browsersteps_start_session['browserstepper'] = browser_steps.browsersteps_live_ui(
+            playwright_browser=browsersteps_start_session['browser'],
+            proxy=proxy)
+
+        # For test
+        #browsersteps_start_session['browserstepper'].action_goto_url(value="http://example.com?time="+str(time.time()))
+
+        return browsersteps_start_session
+
+
     @login_optionally_required
-    @browser_steps_blueprint.route("/browsersteps_update", methods=['GET', 'POST'])
+    @browser_steps_blueprint.route("/browsersteps_start_session", methods=['GET'])
+    def browsersteps_start_session():
+        # A new session was requested, return sessionID
+
+        import uuid
+        browsersteps_session_id = str(uuid.uuid4())
+
+        watch_uuid = request.args.get('uuid')
+        global browsersteps_sessions
+
+        print("Starting connection with playwright")
+        logging.debug("browser_steps.py connecting")
+        browsersteps_sessions[browsersteps_session_id] = start_browsersteps_session()
+        print("Starting connection with playwright - done")
+        return {'browsersteps_session_id': browsersteps_session_id}
+
+    # A request for an action was received
+    @login_optionally_required
+    @browser_steps_blueprint.route("/browsersteps_update", methods=['POST'])
     def browsersteps_ui_update():
         import base64
         import playwright._impl._api_types
-        import time
-
+        global browsersteps_sessions
         from changedetectionio.blueprint.browser_steps import browser_steps
 
-        global browsersteps_live_ui_o, browsersteps_playwright_browser_interface_end_time
-        global browsersteps_playwright_browser_interface_browser
-        global browsersteps_playwright_browser_interface
-        global browsersteps_playwright_browser_interface_start_time
-
-        step_n = None
         remaining =0
         uuid = request.args.get('uuid')
 
@@ -87,13 +132,9 @@ def construct_blueprint(datastore: ChangeDetectionStore):
         if not browsersteps_session_id:
             return make_response('No browsersteps_session_id specified', 500)
 
-        # Because we don't "really" run in a context manager ( we make the playwright interface global/long-living )
-        # We need to manage the shutdown when the time is up
-        if browsersteps_playwright_browser_interface_end_time:
-            remaining = browsersteps_playwright_browser_interface_end_time-time.time()
-            if browsersteps_playwright_browser_interface_end_time and remaining <= 0:
-                cleanup_playwright_session()
-                return make_response('Browser session expired, please reload the Browser Steps interface', 401)
+        if not browsersteps_sessions.get(browsersteps_session_id):
+            return make_response('No session exists under that ID', 500)
+
 
         # Actions - step/apply/etc, do the thing and return state
         if request.method == 'POST':
@@ -112,12 +153,7 @@ def construct_blueprint(datastore: ChangeDetectionStore):
             # @todo try.. accept.. nice errors not popups..
             try:
 
-                this_session = browsersteps_live_ui_o.get(browsersteps_session_id)
-                if not this_session:
-                    print("Browser exited")
-                    return make_response('Browser session ran out of time :( Please reload this page.', 401)
-
-                this_session.call_action(action_name=step_operation,
+                browsersteps_sessions[browsersteps_session_id]['browserstepper'].call_action(action_name=step_operation,
                                          selector=step_selector,
                                          optional_value=step_optional_value)
 
@@ -129,108 +165,43 @@ def construct_blueprint(datastore: ChangeDetectionStore):
             # Get visual selector ready/update its data (also use the current filter info from the page?)
             # When the last 'apply' button was pressed
             # @todo this adds overhead because the xpath selection is happening twice
-            u = this_session.page.url
+            u = browsersteps_sessions[browsersteps_session_id]['browserstepper'].page.url
             if is_last_step and u:
-                (screenshot, xpath_data) = this_session.request_visualselector_data()
+                (screenshot, xpath_data) = browsersteps_sessions[browsersteps_session_id]['browserstepper'].request_visualselector_data()
                 datastore.save_screenshot(watch_uuid=uuid, screenshot=screenshot)
                 datastore.save_xpath_data(watch_uuid=uuid, data=xpath_data)
 
-        # Setup interface
-        if request.method == 'GET':
+#        if not this_session.page:
+#            cleanup_playwright_session()
+#            return make_response('Browser session ran out of time :( Please reload this page.', 401)
 
-            if not browsersteps_playwright_browser_interface:
-                print("Starting connection with playwright")
-                logging.debug("browser_steps.py connecting")
+        # Screenshots and other info only needed on requesting a step (POST)
+        try:
+            state = browsersteps_sessions[browsersteps_session_id]['browserstepper'].get_current_state()
+        except playwright._impl._api_types.Error as e:
+            return make_response("Browser session ran out of time :( Please reload this page."+str(e), 401)
 
-                global browsersteps_playwright_browser_interface_context
-                from . import nonContext
-                browsersteps_playwright_browser_interface_context = nonContext.c_sync_playwright()
-                browsersteps_playwright_browser_interface = browsersteps_playwright_browser_interface_context.start()
-                # At 20 minutes, some other variable is closing it
-                # @todo find out what it is and set it
-                seconds_keepalive = int(os.getenv('BROWSERSTEPS_MINUTES_KEEPALIVE', 10)) * 60
+        # Use send_file() which is way faster than read/write loop on bytes
+        import json
+        from tempfile import mkstemp
+        from flask import send_file
+        tmp_fd, tmp_file = mkstemp(text=True, suffix=".json", prefix="changedetectionio-")
 
-                # keep it alive for 10 seconds more than we advertise, sometimes it helps to keep it shutting down cleanly
-                keepalive = "&timeout={}".format(((seconds_keepalive+3) * 1000))
-                try:
-                    browsersteps_playwright_browser_interface_browser = browsersteps_playwright_browser_interface.chromium.connect_over_cdp(
-                        os.getenv('PLAYWRIGHT_DRIVER_URL', '') + keepalive)
-                except Exception as e:
-                    if 'ECONNREFUSED' in str(e):
-                        return make_response('Unable to start the Playwright session properly, is it running?', 401)
+        output = json.dumps({'screenshot': "data:image/jpeg;base64,{}".format(
+            base64.b64encode(state[0]).decode('ascii')),
+            'xpath_data': state[1],
+            'session_age_start': browsersteps_sessions[browsersteps_session_id]['browserstepper'].age_start,
+            'browser_time_remaining': round(remaining)
+        })
 
-                browsersteps_playwright_browser_interface_end_time = time.time() + (seconds_keepalive-3)
-                print("Starting connection with playwright - done")
+        with os.fdopen(tmp_fd, 'w') as f:
+            f.write(output)
 
-            if not browsersteps_live_ui_o.get(browsersteps_session_id):
-                # Boot up a new session
-                proxy_id = datastore.get_preferred_proxy_for_watch(uuid=uuid)
-                proxy = None
-                if proxy_id:
-                    proxy_url = datastore.proxy_list.get(proxy_id).get('url')
-                    if proxy_url:
-
-                        # Playwright needs separate username and password values
-                        from urllib.parse import urlparse
-                        parsed = urlparse(proxy_url)
-                        proxy = {'server': proxy_url}
-
-                        if parsed.username:
-                            proxy['username'] = parsed.username
-
-                        if parsed.password:
-                            proxy['password'] = parsed.password
-
-                        print("Browser Steps: UUID {} Using proxy {}".format(uuid, proxy_url))
-
-                # Begin the new "Playwright Context" that re-uses the playwright interface
-                # Each session is a "Playwright Context" as a list, that uses the playwright interface
-                browsersteps_live_ui_o[browsersteps_session_id] = browser_steps.browsersteps_live_ui(
-                    playwright_browser=browsersteps_playwright_browser_interface_browser,
-                    proxy=proxy)
-                this_session = browsersteps_live_ui_o[browsersteps_session_id]
-
-        if not this_session.page:
-            cleanup_playwright_session()
-            return make_response('Browser session ran out of time :( Please reload this page.', 401)
-
-        response = None
-
-        if request.method == 'POST':
-            # Screenshots and other info only needed on requesting a step (POST)
-            try:
-                state = this_session.get_current_state()
-            except playwright._impl._api_types.Error as e:
-                return make_response("Browser session ran out of time :( Please reload this page."+str(e), 401)
-
-            # Use send_file() which is way faster than read/write loop on bytes
-            import json
-            from tempfile import mkstemp
-            from flask import send_file
-            tmp_fd, tmp_file = mkstemp(text=True, suffix=".json", prefix="changedetectionio-")
-
-            output = json.dumps({'screenshot': "data:image/jpeg;base64,{}".format(
-                base64.b64encode(state[0]).decode('ascii')),
-                'xpath_data': state[1],
-                'session_age_start': this_session.age_start,
-                'browser_time_remaining': round(remaining)
-            })
-
-            with os.fdopen(tmp_fd, 'w') as f:
-                f.write(output)
-
-            response = make_response(send_file(path_or_file=tmp_file,
-                                               mimetype='application/json; charset=UTF-8',
-                                               etag=True))
-            # No longer needed
-            os.unlink(tmp_file)
-
-        elif request.method == 'GET':
-            # Just enough to get the session rolling, it will call for goto-site via POST next
-            response = make_response({
-                'session_age_start': this_session.age_start,
-                'browser_time_remaining': round(remaining)
-            })
+        response = make_response(send_file(path_or_file=tmp_file,
+                                           mimetype='application/json; charset=UTF-8',
+                                           etag=True))
+        # No longer needed
+        os.unlink(tmp_file)
 
         return response
 

changedetectionio/blueprint/browser_steps/browser_steps.py

@@ -71,10 +71,10 @@ class steppable_browser_interface():
             optional_value = str(jinja2_env.from_string(optional_value).render())
 
         action_handler(selector, optional_value)
-        self.page.wait_for_timeout(3 * 1000)
+        self.page.wait_for_timeout(1.5 * 1000)
         print("Call action done in", time.time() - now)
 
-    def action_goto_url(self, selector, value):
+    def action_goto_url(self, selector=None, value=None):
         # self.page.set_viewport_size({"width": 1280, "height": 5000})
         now = time.time()
         response = self.page.goto(value, timeout=0, wait_until='commit')
@@ -105,7 +105,8 @@ class steppable_browser_interface():
         print("Clicking element")
         if not len(selector.strip()):
             return
-        self.page.click(selector, timeout=10 * 1000, delay=randint(200, 500))
+
+        self.page.click(selector=selector, timeout=30 * 1000, delay=randint(200, 500))
 
     def action_click_element_if_exists(self, selector, value):
         import playwright._impl._api_types as _api_types
@@ -137,13 +138,13 @@ class steppable_browser_interface():
     def action_wait_for_text(self, selector, value):
         import json
         v = json.dumps(value)
-        self.page.wait_for_function(f'document.querySelector("body").innerText.includes({v});', timeout=30000)
+        self.page.wait_for_function(f'document.querySelector("body").innerText.includes({v});', timeout=90000)
 
     def action_wait_for_text_in_element(self, selector, value):
         import json
         s = json.dumps(selector)
         v = json.dumps(value)
-        self.page.wait_for_function(f'document.querySelector({s}).innerText.includes({v});', timeout=30000)
+        self.page.wait_for_function(f'document.querySelector({s}).innerText.includes({v});', timeout=90000)
 
     # @todo - in the future make some popout interface to capture what needs to be set
     # https://playwright.dev/python/docs/api/class-keyboard

changedetectionio/content_fetcher.py

@@ -182,7 +182,8 @@ class Fetcher():
                                                       optional_value=optional_value)
                     self.screenshot_step(step_n)
                     self.save_step_html(step_n)
-                except TimeoutError:
+                except TimeoutError as e:
+                    print(str(e))
                     # Stop processing here
                     raise BrowserStepsStepTimout(step_n=step_n)
 
@@ -287,168 +288,18 @@ class base_html_playwright(Fetcher):
             current_include_filters=None,
             is_binary=False):
 
+        from pkg_resources import resource_string
+
         extra_wait_ms = (int(os.getenv("WEBDRIVER_DELAY_BEFORE_CONTENT_READY", 5)) + self.render_extract_delay) * 1000
-        xpath_element_js = self.xpath_element_js.replace('%ELEMENTS%', visualselector_xpath_selectors)
 
-        code = f"""module.exports = async ({{ page, context }}) => {{
-        
-          var {{ url, execute_js, user_agent, extra_wait_ms, req_headers, include_filters, xpath_element_js, screenshot_quality, proxy_username, proxy_password, disk_cache_dir}} = context;
-          
-          await page.setBypassCSP(true)
-          await page.setExtraHTTPHeaders(req_headers);          
-          await page.setUserAgent(user_agent);
-          // https://ourcodeworld.com/articles/read/1106/how-to-solve-puppeteer-timeouterror-navigation-timeout-of-30000-ms-exceeded
-          
-          await page.setDefaultNavigationTimeout(0);
-
-          if(proxy_username) {{
-            await page.authenticate({{
-                username: proxy_username,
-                password: proxy_password
-            }});
-          }}
-
-        await page.setViewport({{
-          width: 1024,
-          height: 768,
-          deviceScaleFactor: 1,
-        }});
-
-        // Very primitive disk cache - USE WITH EXTREME CAUTION
-        // Run browserless container with -e "FUNCTION_BUILT_INS=[\"fs\",\"crypto\"]"
-        if ( disk_cache_dir ) {{
-            
-            await page.setRequestInterception(true);
-                         
-            console.log(">>>>>>>>>>>>>>> LOCAL DISK CACHE ENABLED <<<<<<<<<<<<<<<<<<<<<");                 
-            const fs = require('fs');
-            const crypto = require('crypto');
-            function file_is_expired(file_path) {{
-                if (!fs.existsSync(dir_path+key)) {{
-                  return true;
-                }}
-                var stats = fs.statSync(file_path);
-                const now_date = new Date();
-                const expire_seconds = 300;
-                if ( (now_date/1000) - (stats.mtime.getTime() / 1000) > expire_seconds) {{                  
-                  console.log("CACHE EXPIRED: "+file_path);
-                  return true;
-                }}
-                return false;
-                
-            }}
-        
-            page.on('request', async (request) => {{
-                    
-                // if (blockedExtensions.some((str) => req.url().endsWith(str))) return req.abort();
-		        const url = request.url();
-                const key = crypto.createHash('md5').update(url).digest("hex");                
-                const dir_path = disk_cache_dir + key.slice(0, 1) + '/' + key.slice(1, 2) + '/' + key.slice(2, 3) + '/';             
-                                       
-                // https://stackoverflow.com/questions/4482686/check-synchronously-if-file-directory-exists-in-node-js
-                
-                if (fs.existsSync(dir_path+key)) {{
-                    file_is_expired(dir_path+key);
-                    console.log("Cache exists "+dir_path+key+ " - "+url);
-                    const cached_data = fs.readFileSync(dir_path+key);                          
-                    request.respond({{
-                        status: 200,
-                        //contentType: 'text/html', //@todo
-                        body: cached_data
-                    }});
-                    return;
-                }}                
-                request.continue();
-            }});
-            
-            page.on('response', async (response) => {{
-                const url = response.url();
-                // @todo - check response size()
-                console.log("Cache - Got "+response.request().method()+" - "+url+" - "+response.request().resourceType());
-                
-                if(response.request().method()  != 'GET' || response.request().resourceType() == 'xhr' || response.request().resourceType() == 'document' || response.status() != 200 ) {{
-                    console.log("Skipping- "+url);
-                    return;
-                }}
-                
-                const key = crypto.createHash('md5').update(url).digest("hex");
-                const dir_path = disk_cache_dir + key.slice(0, 1) + '/' + key.slice(1, 2) + '/' + key.slice(2, 3) + '/';               
-                const data = await response.text();
-                if (!fs.existsSync(dir_path)) {{
-                    fs.mkdirSync(dir_path, {{ recursive: true }})
-                }}
-                
-                var expired = false;
-                if (fs.existsSync(dir_path+key)) {{
-                  if (file_is_expired(dir_path+key)) {{
-                    fs.writeFileSync(dir_path+key, data);
-                  }}
-                }} else {{                
-                    fs.writeFileSync(dir_path+key, data);
-                }}
-		    }});		    
-          }}
-
-        
-          const r = await page.goto(url, {{
-                waitUntil: 'load'                
-          }});
-                            
-          await page.waitForTimeout(1000); 
-          await page.waitForTimeout(extra_wait_ms);
-          
-          if(execute_js) {{
-            await page.evaluate(execute_js);
-            await page.waitForTimeout(200);
-          }}
-          
-        var xpath_data;
-        var instock_data;
-        try {{
-             xpath_data = await page.evaluate((include_filters) => {{ {xpath_element_js} }}, include_filters);
-             instock_data = await page.evaluate(() => {{ {self.instock_data_js} }});
-        }} catch (e) {{
-            console.log(e);
-        }}   
-          
-      // Protocol error (Page.captureScreenshot): Cannot take screenshot with 0 width can come from a proxy auth failure
-      // Wrap it here (for now)
-      
-      var b64s = false;
-      try {{      
-             b64s = await page.screenshot({{ encoding: "base64", fullPage: true, quality: screenshot_quality, type: 'jpeg' }});
-        }} catch (e) {{
-            console.log(e);
-        }}
-        
-        // May fail on very large pages with 'WARNING: tile memory limits exceeded, some content may not draw'
-        if (!b64s) {{
-            // @todo after text extract, we can place some overlay text with red background to say 'croppped'        
-            console.error('ERROR: content-fetcher page was maybe too large for a screenshot, reverting to viewport only screenshot');
-            try {{
-                 b64s = await page.screenshot({{ encoding: "base64", quality: screenshot_quality, type: 'jpeg' }});
-            }} catch (e) {{
-                console.log(e);
-            }}
-         }}
-    
-            
-         var html = await page.content();
-          return {{
-            data: {{
-                'content': html, 
-                'headers': r.headers(), 
-                'instock_data': instock_data,
-                'screenshot': b64s,
-                'status_code': r.status(),
-                'xpath_data': xpath_data
-            }},
-            type: 'application/json',
-          }};
-        }};"""
+        self.xpath_element_js = self.xpath_element_js.replace('%ELEMENTS%', visualselector_xpath_selectors)
+        code = resource_string(__name__, "res/puppeteer_fetch.js").decode('utf-8')
+        # In the future inject this is a proper JS package
+        code = code.replace('%xpath_scrape_code%', self.xpath_element_js)
+        code = code.replace('%instock_scrape_code%', self.instock_data_js)
 
         from requests.exceptions import ConnectTimeout, ReadTimeout
-        wait_browserless_seconds = 120
+        wait_browserless_seconds = 240
 
         browserless_function_url = os.getenv('BROWSERLESS_FUNCTION_URL')
         from urllib.parse import urlparse
@@ -475,7 +326,9 @@ class base_html_playwright(Fetcher):
                 json={
                     "code": code,
                     "context": {
-                        'disk_cache_dir': False, # or path to disk cache
+                        # Very primitive disk cache - USE WITH EXTREME CAUTION
+                        # Run browserless container  with -e "FUNCTION_BUILT_INS=[\"fs\",\"crypto\"]"
+                        'disk_cache_dir': os.getenv("PUPPETEER_DISK_CACHE", False), # or path to disk cache ending in /, ie /tmp/cache/
                         'execute_js': self.webdriver_js_execute_code,
                         'extra_wait_ms': extra_wait_ms,
                         'include_filters': current_include_filters,
@@ -484,14 +337,26 @@ class base_html_playwright(Fetcher):
                         'url': url,
                         'user_agent': request_headers.get('User-Agent', 'Mozilla/5.0'),
                         'proxy_username': self.proxy.get('username','') if self.proxy else False,
-                        'proxy_password': self.proxy.get('password','') if self.proxy else False,
+                        'proxy_password': self.proxy.get('password', '') if self.proxy else False,
+                        'no_cache_list': [
+                            'twitter',
+                            '.pdf'
+                        ],
+                        # Could use https://github.com/easylist/easylist here, or install a plugin
+                        'block_url_list': [
+                            'adnxs.com',
+                            'analytics.twitter.com',
+                            'doubleclick.net',
+                            'google-analytics.com',
+                            'googletagmanager',
+                            'trustpilot.com'
+                        ]
                     }
                 },
                 # @todo /function needs adding ws:// to http:// rebuild this
                 url=browserless_function_url+f"{amp}--disable-features=AudioServiceOutOfProcess&dumpio=true&--disable-remote-fonts",
                 timeout=wait_browserless_seconds)
 
-# 'ziparchive::addglob() will throw an instance of error instead of resulting in a fatal error if glob support is not available.'
         except ReadTimeout:
             raise PageUnloadable(url=url, status_code=None, message=f"No response from browserless in {wait_browserless_seconds}s")
         except ConnectTimeout:
@@ -535,17 +400,23 @@ class base_html_playwright(Fetcher):
             current_include_filters=None,
             is_binary=False):
 
-        if os.getenv('USE_EXPERIMENTAL_PUPPETEER_FETCH'):
-            # Temporary backup solution until we rewrite the playwright code
-            return self.run_fetch_browserless_puppeteer(
-                url,
-                timeout,
-                request_headers,
-                request_body,
-                request_method,
-                ignore_status_codes,
-                current_include_filters,
-                is_binary)
+        # For now, USE_EXPERIMENTAL_PUPPETEER_FETCH is not supported by watches with BrowserSteps (for now!)
+        has_browser_steps = self.browser_steps and list(filter(
+                lambda s: (s['operation'] and len(s['operation']) and s['operation'] != 'Choose one' and s['operation'] != 'Goto site'),
+                self.browser_steps))
+
+        if not has_browser_steps:
+            if os.getenv('USE_EXPERIMENTAL_PUPPETEER_FETCH'):
+                # Temporary backup solution until we rewrite the playwright code
+                return self.run_fetch_browserless_puppeteer(
+                    url,
+                    timeout,
+                    request_headers,
+                    request_body,
+                    request_method,
+                    ignore_status_codes,
+                    current_include_filters,
+                    is_binary)
 
         from playwright.sync_api import sync_playwright
         import playwright._impl._api_types

changedetectionio/res/puppeteer_fetch.js

@@ -0,0 +1,183 @@
+module.exports = async ({page, context}) => {
+
+    var {
+        url,
+        execute_js,
+        user_agent,
+        extra_wait_ms,
+        req_headers,
+        include_filters,
+        xpath_element_js,
+        screenshot_quality,
+        proxy_username,
+        proxy_password,
+        disk_cache_dir,
+        no_cache_list,
+        block_url_list,
+    } = context;
+
+    await page.setBypassCSP(true)
+    await page.setExtraHTTPHeaders(req_headers);
+    await page.setUserAgent(user_agent);
+    // https://ourcodeworld.com/articles/read/1106/how-to-solve-puppeteer-timeouterror-navigation-timeout-of-30000-ms-exceeded
+
+    await page.setDefaultNavigationTimeout(0);
+
+    if (proxy_username) {
+        await page.authenticate({
+            username: proxy_username,
+            password: proxy_password
+        });
+    }
+
+    await page.setViewport({
+        width: 1024,
+        height: 768,
+        deviceScaleFactor: 1,
+    });
+
+    await page.setRequestInterception(true);
+    if (disk_cache_dir) {
+        console.log(">>>>>>>>>>>>>>> LOCAL DISK CACHE ENABLED <<<<<<<<<<<<<<<<<<<<<");
+    }
+    const fs = require('fs');
+    const crypto = require('crypto');
+
+    function file_is_expired(file_path) {
+        if (!fs.existsSync(file_path)) {
+            return true;
+        }
+        var stats = fs.statSync(file_path);
+        const now_date = new Date();
+        const expire_seconds = 300;
+        if ((now_date / 1000) - (stats.mtime.getTime() / 1000) > expire_seconds) {
+            console.log("CACHE EXPIRED: " + file_path);
+            return true;
+        }
+        return false;
+
+    }
+
+    page.on('request', async (request) => {
+        // General blocking of requests that waste traffic
+        if (block_url_list.some(substring => request.url().toLowerCase().includes(substring))) return request.abort();
+
+        if (disk_cache_dir) {
+            const url = request.url();
+            const key = crypto.createHash('md5').update(url).digest("hex");
+            const dir_path = disk_cache_dir + key.slice(0, 1) + '/' + key.slice(1, 2) + '/' + key.slice(2, 3) + '/';
+
+            // https://stackoverflow.com/questions/4482686/check-synchronously-if-file-directory-exists-in-node-js
+
+            if (fs.existsSync(dir_path + key)) {
+                console.log("* CACHE HIT , using - " + dir_path + key + " - " + url);
+                const cached_data = fs.readFileSync(dir_path + key);
+                // @todo headers can come from dir_path+key+".meta" json file
+                request.respond({
+                    status: 200,
+                    //contentType: 'text/html', //@todo
+                    body: cached_data
+                });
+                return;
+            }
+        }
+        request.continue();
+    });
+
+
+    if (disk_cache_dir) {
+        page.on('response', async (response) => {
+            const url = response.url();
+            // Basic filtering for sane responses
+            if (response.request().method() != 'GET' || response.request().resourceType() == 'xhr' || response.request().resourceType() == 'document' || response.status() != 200) {
+                console.log("Skipping (not useful) - Status:" + response.status() + " Method:" + response.request().method() + " ResourceType:" + response.request().resourceType() + " " + url);
+                return;
+            }
+            if (no_cache_list.some(substring => url.toLowerCase().includes(substring))) {
+                console.log("Skipping (no_cache_list) - " + url);
+                return;
+            }
+            if (url.toLowerCase().includes('data:')) {
+                console.log("Skipping (embedded-data) - " + url);
+                return;
+            }
+            response.buffer().then(buffer => {
+                if (buffer.length > 100) {
+                    console.log("Cache - Saving " + response.request().method() + " - " + url + " - " + response.request().resourceType());
+
+                    const key = crypto.createHash('md5').update(url).digest("hex");
+                    const dir_path = disk_cache_dir + key.slice(0, 1) + '/' + key.slice(1, 2) + '/' + key.slice(2, 3) + '/';
+
+                    if (!fs.existsSync(dir_path)) {
+                        fs.mkdirSync(dir_path, {recursive: true})
+                    }
+
+                    if (fs.existsSync(dir_path + key)) {
+                        if (file_is_expired(dir_path + key)) {
+                            fs.writeFileSync(dir_path + key, buffer);
+                        }
+                    } else {
+                        fs.writeFileSync(dir_path + key, buffer);
+                    }
+                }
+            });
+        });
+    }
+
+    const r = await page.goto(url, {
+        waitUntil: 'load'
+    });
+
+    await page.waitForTimeout(1000);
+    await page.waitForTimeout(extra_wait_ms);
+
+    if (execute_js) {
+        await page.evaluate(execute_js);
+        await page.waitForTimeout(200);
+    }
+
+    var xpath_data;
+    var instock_data;
+    try {
+        // Not sure the best way here, in the future this should be a new package added to npm then run in browserless
+        // (Once the old playwright is removed)
+        xpath_data = await page.evaluate((include_filters) => {%xpath_scrape_code%}, include_filters);
+        instock_data = await page.evaluate(() => {%instock_scrape_code%});
+    } catch (e) {
+        console.log(e);
+    }
+
+    // Protocol error (Page.captureScreenshot): Cannot take screenshot with 0 width can come from a proxy auth failure
+    // Wrap it here (for now)
+
+    var b64s = false;
+    try {
+        b64s = await page.screenshot({encoding: "base64", fullPage: true, quality: screenshot_quality, type: 'jpeg'});
+    } catch (e) {
+        console.log(e);
+    }
+
+    // May fail on very large pages with 'WARNING: tile memory limits exceeded, some content may not draw'
+    if (!b64s) {
+        // @todo after text extract, we can place some overlay text with red background to say 'croppped'
+        console.error('ERROR: content-fetcher page was maybe too large for a screenshot, reverting to viewport only screenshot');
+        try {
+            b64s = await page.screenshot({encoding: "base64", quality: screenshot_quality, type: 'jpeg'});
+        } catch (e) {
+            console.log(e);
+        }
+    }
+
+    var html = await page.content();
+    return {
+        data: {
+            'content': html,
+            'headers': r.headers(),
+            'instock_data': instock_data,
+            'screenshot': b64s,
+            'status_code': r.status(),
+            'xpath_data': xpath_data
+        },
+        type: 'application/json',
+    };
+};

changedetectionio/static/js/browser-steps.js

@@ -114,11 +114,11 @@ $(document).ready(function () {
             e.preventDefault()
         });
 
+        // When the mouse moves we know which element it should be above
+        // mousedown will link that to the UI (select the right action, highlight etc)
         $('#browsersteps-selector-canvas').bind('mousedown', function (e) {
             // https://developer.mozilla.org/en-US/docs/Web/API/MouseEvent
             e.preventDefault()
-            console.log(e);
-            console.log("current xpath in index is " + current_selected_i);
             last_click_xy = {'x': parseInt((1 / x_scale) * e.offsetX), 'y': parseInt((1 / y_scale) * e.offsetY)}
             process_selected(current_selected_i);
             current_selected_i = false;
@@ -132,6 +132,7 @@ $(document).ready(function () {
             }
         });
 
+        // Debounce and find the current most 'interesting' element we are hovering above
         $('#browsersteps-selector-canvas').bind('mousemove', function (e) {
             if (!xpath_data) {
                 return;
@@ -151,41 +152,38 @@ $(document).ready(function () {
             current_selected_i = false;
             // Reverse order - the most specific one should be deeper/"laster"
             // Basically, find the most 'deepest'
-            //$('#browsersteps-selector-canvas').css('cursor', 'pointer');
-            for (var i = xpath_data['size_pos'].length; i !== 0; i--) {
-                // draw all of them? let them choose somehow?
-                var sel = xpath_data['size_pos'][i - 1];
+            var possible_elements = [];
+            xpath_data['size_pos'].forEach(function (item, index) {
                 // If we are in a bounding-box
-                if (e.offsetY > sel.top * y_scale && e.offsetY < sel.top * y_scale + sel.height * y_scale
+                if (e.offsetY > item.top * y_scale && e.offsetY < item.top * y_scale + item.height * y_scale
                     &&
-                    e.offsetX > sel.left * y_scale && e.offsetX < sel.left * y_scale + sel.width * y_scale
+                    e.offsetX > item.left * y_scale && e.offsetX < item.left * y_scale + item.width * y_scale
 
                 ) {
-                    // Only highlight these interesting types
-                    if (1) {
-                        ctx.strokeRect(sel.left * x_scale, sel.top * y_scale, sel.width * x_scale, sel.height * y_scale);
-                        ctx.fillRect(sel.left * x_scale, sel.top * y_scale, sel.width * x_scale, sel.height * y_scale);
-                        current_selected_i = i - 1;
-                        break;
-
-                        // find the smallest one at this x,y
-                        // does it mean sort the xpath list by size (w*h) i think so!
-                    } else {
-
-                        if (include_text_elements[0].checked === true) {
-                            // blue one with background instead?
-                            ctx.fillStyle = 'rgba(0,0,255, 0.1)';
-                            ctx.strokeStyle = 'rgba(0,0,200, 0.7)';
-                            $('#browsersteps-selector-canvas').css('cursor', 'grab');
-                            ctx.strokeRect(sel.left * x_scale, sel.top * y_scale, sel.width * x_scale, sel.height * y_scale);
-                            ctx.fillRect(sel.left * x_scale, sel.top * y_scale, sel.width * x_scale, sel.height * y_scale);
-                            current_selected_i = i - 1;
-                            break;
-                        }
-                    }
+                    // There could be many elements here, record them all and then we'll find out which is the most 'useful'
+                    // (input, textarea, button, A etc)
+                    possible_elements.push(item);
                 }
+            });
+
+            // Find the best one
+            if (possible_elements.length) {
+                possible_elements.forEach(function (item, index) {
+                  if (["a", "input", "textarea", "button"].includes(item['tagName'])) {
+                      current_selected_i = item;
+                  }
+                });
+
+                if (!current_selected_i) {
+                    current_selected_i = possible_elements[0];
+                }
+
+                sel = xpath_data['size_pos'][current_selected_i];
+                ctx.strokeRect(current_selected_i.left * x_scale, current_selected_i.top * y_scale, current_selected_i.width * x_scale, current_selected_i.height * y_scale);
+                ctx.fillRect(current_selected_i.left * x_scale, current_selected_i.top * y_scale, current_selected_i.width * x_scale, current_selected_i.height * y_scale);
             }
 
+
         }.debounce(10));
     });
 
@@ -195,16 +193,16 @@ $(document).ready(function () {
 
 
     // callback for clicking on an xpath on the canvas
-    function process_selected(xpath_data_index) {
+    function process_selected(selected_in_xpath_list) {
         found_something = false;
         var first_available = $("ul#browser_steps li.empty").first();
 
 
-        if (xpath_data_index !== false) {
+        if (selected_in_xpath_list !== false) {
             // Nothing focused, so fill in a new one
             // if inpt type button or <button>
             // from the top, find the next not used one and use it
-            var x = xpath_data['size_pos'][xpath_data_index];
+            var x = selected_in_xpath_list;
             console.log(x);
             if (x && first_available.length) {
                 // @todo will it let you click shit that has a layer ontop? probably not.
@@ -214,26 +212,18 @@ $(document).ready(function () {
                     $('input[placeholder="Value"]', first_available).addClass('ok').click().focus();
                     found_something = true;
                 } else {
-                    if (x['isClickable'] || x['tagName'].startsWith('h') || x['tagName'] === 'a' || x['tagName'] === 'button' || x['tagtype'] === 'submit' || x['tagtype'] === 'checkbox' || x['tagtype'] === 'radio' || x['tagtype'] === 'li') {
+                    // There's no good way (that I know) to find if this
+                    // see https://stackoverflow.com/questions/446892/how-to-find-event-listeners-on-a-dom-node-in-javascript-or-in-debugging
+                    // https://codepen.io/azaslavsky/pen/DEJVWv
+
+                    // So we dont know if its really a clickable element or not :-(
+                    // Assume it is - then we dont fill the pages with unreliable "Click X,Y" selections
+                    // If you switch to "Click X,y" after an element here is setup, it will give the last co-ords anyway
+                    //if (x['isClickable'] || x['tagName'].startsWith('h') || x['tagName'] === 'a' || x['tagName'] === 'button' || x['tagtype'] === 'submit' || x['tagtype'] === 'checkbox' || x['tagtype'] === 'radio' || x['tagtype'] === 'li') {
                         $('select', first_available).val('Click element').change();
                         $('input[type=text]', first_available).first().val(x['xpath']);
                         found_something = true;
-                    }
-                }
-
-                first_available.xpath_data_index = xpath_data_index;
-
-                if (!found_something) {
-                    if (include_text_elements[0].checked === true) {
-                        // Suggest that we use as filter?
-                        // @todo filters should always be in the last steps, nothing non-filter after it
-                        found_something = true;
-                        ctx.strokeStyle = 'rgba(0,0,255, 0.9)';
-                        ctx.fillStyle = 'rgba(0,0,255, 0.1)';
-                        $('select', first_available).val('Extract text and use as filter').change();
-                        $('input[type=text]', first_available).first().val(x['xpath']);
-                        include_text_elements[0].checked = false;
-                    }
+                    //}
                 }
             }
         }
@@ -248,7 +238,7 @@ $(document).ready(function () {
 
     function start() {
         console.log("Starting browser-steps UI");
-        browsersteps_session_id = Date.now();
+        browsersteps_session_id = false;
         // @todo This setting of the first one should be done at the datalayer but wtforms doesnt wanna play nice
         $('#browser_steps >li:first-child').removeClass('empty');
         set_first_gotosite_disabled();
@@ -256,7 +246,7 @@ $(document).ready(function () {
         $('.clear,.remove', $('#browser_steps >li:first-child')).hide();
         $.ajax({
             type: "GET",
-            url: browser_steps_sync_url + "&browsersteps_session_id=" + browsersteps_session_id,
+            url: browser_steps_start_url,
             statusCode: {
                 400: function () {
                     // More than likely the CSRF token was lost when the server restarted
@@ -264,12 +254,12 @@ $(document).ready(function () {
                 }
             }
         }).done(function (data) {
-            xpath_data = data.xpath_data;
             $("#loading-status-text").fadeIn();
+            browsersteps_session_id = data.browsersteps_session_id;
             // This should trigger 'Goto site'
             console.log("Got startup response, requesting Goto-Site (first) step fake click");
             $('#browser_steps >li:first-child .apply').click();
-            browserless_seconds_remaining = data.browser_time_remaining;
+            browserless_seconds_remaining = 500;
             set_first_gotosite_disabled();
         }).fail(function (data) {
             console.log(data);

changedetectionio/templates/edit.html

@@ -14,7 +14,9 @@
 {% endif %}
 
     const browser_steps_config=JSON.parse('{{ browser_steps_config|tojson }}');
+    const browser_steps_start_url="{{url_for('browser_steps.browsersteps_start_session', uuid=uuid)}}";
     const browser_steps_sync_url="{{url_for('browser_steps.browsersteps_ui_update', uuid=uuid)}}";
+
 </script>
 
 <script src="{{url_for('static_content', group='js', filename='watch-settings.js')}}" defer></script>
@@ -186,7 +188,8 @@ User-Agent: wonderbra 1.0") }}
                                     <span class="loader" >
                                         <span id="browsersteps-click-start">
                                             <h2 >Click here to Start</h2>
-                                            Please allow 10-15 seconds for the browser to connect.
+                                            <svg style="height: 3.5rem;" version="1.1" viewBox="0 0 32 32"  xml:space="preserve" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink"><g id="Layer_1"/><g id="play_x5F_alt"><path d="M16,0C7.164,0,0,7.164,0,16s7.164,16,16,16s16-7.164,16-16S24.836,0,16,0z M10,24V8l16.008,8L10,24z" style="fill: var(--color-grey-400);"/></g></svg><br>
+                                            Please allow 10-15 seconds for the browser to connect.<br>
                                         </span>
                                         <div class="spinner"  style="display: none;"></div>
                                     </span>