001// License: GPL. For details, see LICENSE file.
002package org.openstreetmap.josm.data.cache;
003
004import java.io.FileNotFoundException;
005import java.io.IOException;
006import java.net.URL;
007import java.security.SecureRandom;
008import java.util.HashSet;
009import java.util.List;
010import java.util.Map;
011import java.util.Set;
012import java.util.concurrent.ConcurrentHashMap;
013import java.util.concurrent.ConcurrentMap;
014import java.util.concurrent.LinkedBlockingDeque;
015import java.util.concurrent.ThreadPoolExecutor;
016import java.util.concurrent.TimeUnit;
017import java.util.logging.Level;
018import java.util.logging.Logger;
019
020import org.apache.commons.jcs.access.behavior.ICacheAccess;
021import org.apache.commons.jcs.engine.behavior.ICacheElement;
022import org.openstreetmap.gui.jmapviewer.FeatureAdapter;
023import org.openstreetmap.josm.Main;
024import org.openstreetmap.josm.data.cache.ICachedLoaderListener.LoadResult;
025import org.openstreetmap.josm.data.preferences.IntegerProperty;
026import org.openstreetmap.josm.tools.HttpClient;
027import org.openstreetmap.josm.tools.Utils;
028
029import sun.net.www.protocol.http.HttpURLConnection;
030
031/**
032 * @author Wiktor Niesiobędzki
033 *
034 * @param <K> cache entry key type
035 * @param <V> cache value type
036 *
037 * Generic loader for HTTP based tiles. Uses custom attribute, to check, if entry has expired
038 * according to HTTP headers sent with tile. If so, it tries to verify using Etags
039 * or If-Modified-Since / Last-Modified.
040 *
041 * If the tile is not valid, it will try to download it from remote service and put it
042 * to cache. If remote server will fail it will try to use stale entry.
043 *
044 * This class will keep only one Job running for specified tile. All others will just finish, but
045 * listeners will be gathered and notified, once download job will be finished
046 *
047 * @since 8168
048 */
049public abstract class JCSCachedTileLoaderJob<K, V extends CacheEntry> implements ICachedLoaderJob<K> {
050    private static final Logger log = FeatureAdapter.getLogger(JCSCachedTileLoaderJob.class.getCanonicalName());
051    protected static final long DEFAULT_EXPIRE_TIME = 1000L * 60 * 60 * 24 * 7; // 7 days
052    // Limit for the max-age value send by the server.
053    protected static final long EXPIRE_TIME_SERVER_LIMIT = 1000L * 60 * 60 * 24 * 28; // 4 weeks
054    // Absolute expire time limit. Cached tiles that are older will not be used,
055    // even if the refresh from the server fails.
056    protected static final long ABSOLUTE_EXPIRE_TIME_LIMIT = 1000L * 60 * 60 * 24 * 365; // 1 year
057
058    /**
059     * maximum download threads that will be started
060     */
061    public static final IntegerProperty THREAD_LIMIT = new IntegerProperty("cache.jcs.max_threads", 10);
062
063    /*
064     * ThreadPoolExecutor starts new threads, until THREAD_LIMIT is reached. Then it puts tasks into LinkedBlockingDeque.
065     *
066     * The queue works FIFO, so one needs to take care about ordering of the entries submitted
067     *
068     * There is no point in canceling tasks, that are already taken by worker threads (if we made so much effort, we can at least cache
069     * the response, so later it could be used). We could actually cancel what is in LIFOQueue, but this is a tradeoff between simplicity
070     * and performance (we do want to have something to offer to worker threads before tasks will be resubmitted by class consumer)
071     */
072
073    private static final ThreadPoolExecutor DEFAULT_DOWNLOAD_JOB_DISPATCHER = new ThreadPoolExecutor(
074            1, // we have a small queue, so threads will be quickly started (threads are started only, when queue is full)
075            THREAD_LIMIT.get(), // do not this number of threads
076            30, // keepalive for thread
077            TimeUnit.SECONDS,
078            // make queue of LIFO type - so recently requested tiles will be loaded first (assuming that these are which user is waiting to see)
079            new LinkedBlockingDeque<Runnable>(),
080            Utils.newThreadFactory("JCS-downloader-%d", Thread.NORM_PRIORITY)
081            );
082
083
084
085    private static final ConcurrentMap<String, Set<ICachedLoaderListener>> inProgress = new ConcurrentHashMap<>();
086    private static final ConcurrentMap<String, Boolean> useHead = new ConcurrentHashMap<>();
087
088    protected final long now; // when the job started
089
090    private final ICacheAccess<K, V> cache;
091    private ICacheElement<K, V> cacheElement;
092    protected V cacheData;
093    protected CacheEntryAttributes attributes;
094
095    // HTTP connection parameters
096    private final int connectTimeout;
097    private final int readTimeout;
098    private final Map<String, String> headers;
099    private final ThreadPoolExecutor downloadJobExecutor;
100    private Runnable finishTask;
101    private boolean force;
102
103    /**
104     * @param cache cache instance that we will work on
105     * @param headers HTTP headers to be sent together with request
106     * @param readTimeout when connecting to remote resource
107     * @param connectTimeout when connecting to remote resource
108     * @param downloadJobExecutor that will be executing the jobs
109     */
110    public JCSCachedTileLoaderJob(ICacheAccess<K, V> cache,
111            int connectTimeout, int readTimeout,
112            Map<String, String> headers,
113            ThreadPoolExecutor downloadJobExecutor) {
114
115        this.cache = cache;
116        this.now = System.currentTimeMillis();
117        this.connectTimeout = connectTimeout;
118        this.readTimeout = readTimeout;
119        this.headers = headers;
120        this.downloadJobExecutor = downloadJobExecutor;
121    }
122
123    /**
124     * @param cache cache instance that we will work on
125     * @param headers HTTP headers to be sent together with request
126     * @param readTimeout when connecting to remote resource
127     * @param connectTimeout when connecting to remote resource
128     */
129    public JCSCachedTileLoaderJob(ICacheAccess<K, V> cache,
130            int connectTimeout, int readTimeout,
131            Map<String, String> headers) {
132        this(cache, connectTimeout, readTimeout,
133                headers, DEFAULT_DOWNLOAD_JOB_DISPATCHER);
134    }
135
136    private void ensureCacheElement() {
137        if (cacheElement == null && getCacheKey() != null) {
138            cacheElement = cache.getCacheElement(getCacheKey());
139            if (cacheElement != null) {
140                attributes = (CacheEntryAttributes) cacheElement.getElementAttributes();
141                cacheData = cacheElement.getVal();
142            }
143        }
144    }
145
146    @Override
147    public V get() {
148        ensureCacheElement();
149        return cacheData;
150    }
151
152    @Override
153    public void submit(ICachedLoaderListener listener, boolean force) throws IOException {
154        this.force = force;
155        boolean first = false;
156        URL url = getUrl();
157        String deduplicationKey = null;
158        if (url != null) {
159            // url might be null, for example when Bing Attribution is not loaded yet
160            deduplicationKey = url.toString();
161        }
162        if (deduplicationKey == null) {
163            log.log(Level.WARNING, "No url returned for: {0}, skipping", getCacheKey());
164            throw new IllegalArgumentException("No url returned");
165        }
166        synchronized (inProgress) {
167            Set<ICachedLoaderListener> newListeners = inProgress.get(deduplicationKey);
168            if (newListeners == null) {
169                newListeners = new HashSet<>();
170                inProgress.put(deduplicationKey, newListeners);
171                first = true;
172            }
173            newListeners.add(listener);
174        }
175
176        if (first || force) {
177            // submit all jobs to separate thread, so calling thread is not blocked with IO when loading from disk
178            log.log(Level.FINE, "JCS - Submitting job for execution for url: {0}", getUrlNoException());
179            downloadJobExecutor.execute(this);
180        }
181    }
182
183    /**
184     * This method is run when job has finished
185     */
186    protected void executionFinished() {
187        if (finishTask != null) {
188            finishTask.run();
189        }
190    }
191
192    /**
193     *
194     * @return checks if object from cache has sufficient data to be returned
195     */
196    protected boolean isObjectLoadable() {
197        if (cacheData == null) {
198            return false;
199        }
200        byte[] content = cacheData.getContent();
201        return content != null && content.length > 0;
202    }
203
204    /**
205     * Simple implementation. All errors should be cached as empty. Though some JDK (JDK8 on Windows for example)
206     * doesn't return 4xx error codes, instead they do throw an FileNotFoundException or IOException
207     *
208     * @return true if we should put empty object into cache, regardless of what remote resource has returned
209     */
210    protected boolean cacheAsEmpty() {
211        return attributes.getResponseCode() < 500;
212    }
213
214    /**
215     * @return key under which discovered server settings will be kept
216     */
217    protected String getServerKey() {
218        return getUrlNoException().getHost();
219    }
220
221    @Override
222    public void run() {
223        final Thread currentThread = Thread.currentThread();
224        final String oldName = currentThread.getName();
225        currentThread.setName("JCS Downloading: " + getUrlNoException());
226        log.log(Level.FINE, "JCS - starting fetch of url: {0} ", getUrlNoException());
227        ensureCacheElement();
228        try {
229            // try to fetch from cache
230            if (!force && cacheElement != null && isCacheElementValid() && isObjectLoadable()) {
231                // we got something in cache, and it's valid, so lets return it
232                log.log(Level.FINE, "JCS - Returning object from cache: {0}", getCacheKey());
233                finishLoading(LoadResult.SUCCESS);
234                return;
235            }
236
237            // try to load object from remote resource
238            if (loadObject()) {
239                finishLoading(LoadResult.SUCCESS);
240            } else {
241                // if loading failed - check if we can return stale entry
242                if (isObjectLoadable()) {
243                    // try to get stale entry in cache
244                    finishLoading(LoadResult.SUCCESS);
245                    log.log(Level.FINE, "JCS - found stale object in cache: {0}", getUrlNoException());
246                } else {
247                    // failed completely
248                    finishLoading(LoadResult.FAILURE);
249                }
250            }
251        } finally {
252            executionFinished();
253            currentThread.setName(oldName);
254        }
255    }
256
257    private void finishLoading(LoadResult result) {
258        Set<ICachedLoaderListener> listeners;
259        synchronized (inProgress) {
260            listeners = inProgress.remove(getUrlNoException().toString());
261        }
262        if (listeners == null) {
263            log.log(Level.WARNING, "Listener not found for URL: {0}. Listener not notified!", getUrlNoException());
264            return;
265        }
266        for (ICachedLoaderListener l: listeners) {
267            l.loadingFinished(cacheData, attributes, result);
268        }
269    }
270
271    protected boolean isCacheElementValid() {
272        long expires = attributes.getExpirationTime();
273
274        // check by expire date set by server
275        if (expires != 0L) {
276            // put a limit to the expire time (some servers send a value
277            // that is too large)
278            expires = Math.min(expires, attributes.getCreateTime() + EXPIRE_TIME_SERVER_LIMIT);
279            if (now > expires) {
280                log.log(Level.FINE, "JCS - Object {0} has expired -> valid to {1}, now is: {2}",
281                        new Object[]{getUrlNoException(), Long.toString(expires), Long.toString(now)});
282                return false;
283            }
284        } else if (attributes.getLastModification() > 0 &&
285                now - attributes.getLastModification() > DEFAULT_EXPIRE_TIME) {
286            // check by file modification date
287            log.log(Level.FINE, "JCS - Object has expired, maximum file age reached {0}", getUrlNoException());
288            return false;
289        } else if (now - attributes.getCreateTime() > DEFAULT_EXPIRE_TIME) {
290            log.log(Level.FINE, "JCS - Object has expired, maximum time since object creation reached {0}", getUrlNoException());
291            return false;
292        }
293        return true;
294    }
295
296    /**
297     * @return true if object was successfully downloaded, false, if there was a loading failure
298     */
299    private boolean loadObject() {
300        if (attributes == null) {
301            attributes = new CacheEntryAttributes();
302        }
303        try {
304            // if we have object in cache, and host doesn't support If-Modified-Since nor If-None-Match
305            // then just use HEAD request and check returned values
306            if (isObjectLoadable() &&
307                    Boolean.TRUE.equals(useHead.get(getServerKey())) &&
308                    isCacheValidUsingHead()) {
309                log.log(Level.FINE, "JCS - cache entry verified using HEAD request: {0}", getUrl());
310                return true;
311            }
312
313            final HttpClient request = getRequest("GET", true);
314
315            if (isObjectLoadable() &&
316                    (now - attributes.getLastModification()) <= ABSOLUTE_EXPIRE_TIME_LIMIT) {
317                request.setIfModifiedSince(attributes.getLastModification());
318            }
319            if (isObjectLoadable() && attributes.getEtag() != null) {
320                request.setHeader("If-None-Match", attributes.getEtag());
321            }
322
323            final HttpClient.Response urlConn = request.connect();
324
325            if (urlConn.getResponseCode() == 304) {
326                // If isModifiedSince or If-None-Match has been set
327                // and the server answers with a HTTP 304 = "Not Modified"
328                log.log(Level.FINE, "JCS - If-Modified-Since/ETag test: local version is up to date: {0}", getUrl());
329                return true;
330            } else if (isObjectLoadable() // we have an object in cache, but we haven't received 304 response code
331                    && (
332                            (attributes.getEtag() != null && attributes.getEtag().equals(urlConn.getHeaderField("ETag"))) ||
333                            attributes.getLastModification() == urlConn.getLastModified())
334                    ) {
335                // we sent ETag or If-Modified-Since, but didn't get 304 response code
336                // for further requests - use HEAD
337                String serverKey = getServerKey();
338                log.log(Level.INFO, "JCS - Host: {0} found not to return 304 codes for If-Modified-Since or If-None-Match headers",
339                        serverKey);
340                useHead.put(serverKey, Boolean.TRUE);
341            }
342
343            attributes = parseHeaders(urlConn);
344
345            for (int i = 0; i < 5; ++i) {
346                if (urlConn.getResponseCode() == HttpURLConnection.HTTP_UNAVAILABLE) {
347                    Thread.sleep(5000L+new SecureRandom().nextInt(5000));
348                    continue;
349                }
350
351                attributes.setResponseCode(urlConn.getResponseCode());
352                byte[] raw;
353                if (urlConn.getResponseCode() == HttpURLConnection.HTTP_OK) {
354                    raw = Utils.readBytesFromStream(urlConn.getContent());
355                } else {
356                    raw = new byte[]{};
357                }
358
359                if (isResponseLoadable(urlConn.getHeaderFields(), urlConn.getResponseCode(), raw)) {
360                    // we need to check cacheEmpty, so for cases, when data is returned, but we want to store
361                    // as empty (eg. empty tile images) to save some space
362                    cacheData = createCacheEntry(raw);
363                    cache.put(getCacheKey(), cacheData, attributes);
364                    log.log(Level.FINE, "JCS - downloaded key: {0}, length: {1}, url: {2}",
365                            new Object[] {getCacheKey(), raw.length, getUrl()});
366                    return true;
367                } else if (cacheAsEmpty()) {
368                    cacheData = createCacheEntry(new byte[]{});
369                    cache.put(getCacheKey(), cacheData, attributes);
370                    log.log(Level.FINE, "JCS - Caching empty object {0}", getUrl());
371                    return true;
372                } else {
373                    log.log(Level.FINE, "JCS - failure during load - reponse is not loadable nor cached as empty");
374                    return false;
375                }
376            }
377        } catch (FileNotFoundException e) {
378            log.log(Level.FINE, "JCS - Caching empty object as server returned 404 for: {0}", getUrlNoException());
379            attributes.setResponseCode(404);
380            attributes.setError(e);
381            boolean doCache = isResponseLoadable(null, 404, null) || cacheAsEmpty();
382            if (doCache) {
383                cacheData = createCacheEntry(new byte[]{});
384                cache.put(getCacheKey(), cacheData, attributes);
385            }
386            return doCache;
387        } catch (IOException e) {
388            log.log(Level.FINE, "JCS - IOExecption during communication with server for: {0}", getUrlNoException());
389            attributes.setError(e);
390            attributes.setResponseCode(499); // set dummy error code
391            boolean doCache = isResponseLoadable(null, 499, null) || cacheAsEmpty(); //generic 499 error code returned
392            if (doCache) {
393                cacheData = createCacheEntry(new byte[]{});
394                cache.put(getCacheKey(), createCacheEntry(new byte[]{}), attributes);
395            }
396            return doCache;
397        } catch (InterruptedException e) {
398            attributes.setError(e);
399            log.log(Level.WARNING, "JCS - Exception during download {0}", getUrlNoException());
400            Main.warn(e);
401        }
402        log.log(Level.WARNING, "JCS - Silent failure during download: {0}", getUrlNoException());
403        return false;
404    }
405
406    /**
407     * Check if the object is loadable. This means, if the data will be parsed, and if this response
408     * will finish as successful retrieve.
409     *
410     * This simple implementation doesn't load empty response, nor client (4xx) and server (5xx) errors
411     *
412     * @param headerFields headers sent by server
413     * @param responseCode http status code
414     * @param raw data read from server
415     * @return true if object should be cached and returned to listener
416     */
417    protected boolean isResponseLoadable(Map<String, List<String>> headerFields, int responseCode, byte[] raw) {
418        if (raw == null || raw.length == 0 || responseCode >= 400) {
419            return false;
420        }
421        return true;
422    }
423
424    protected abstract V createCacheEntry(byte[] content);
425
426    protected CacheEntryAttributes parseHeaders(HttpClient.Response urlConn) {
427        CacheEntryAttributes ret = new CacheEntryAttributes();
428
429        Long lng = urlConn.getExpiration();
430        if (lng.equals(0L)) {
431            try {
432                String str = urlConn.getHeaderField("Cache-Control");
433                if (str != null) {
434                    for (String token: str.split(",")) {
435                        if (token.startsWith("max-age=")) {
436                            lng = Long.parseLong(token.substring(8)) * 1000 +
437                                    System.currentTimeMillis();
438                        }
439                    }
440                }
441            } catch (NumberFormatException e) {
442                // ignore malformed Cache-Control headers
443                if (Main.isTraceEnabled()) {
444                    Main.trace(e.getMessage());
445                }
446            }
447        }
448
449        ret.setExpirationTime(lng);
450        ret.setLastModification(now);
451        ret.setEtag(urlConn.getHeaderField("ETag"));
452
453        return ret;
454    }
455
456    private HttpClient getRequest(String requestMethod, boolean noCache) throws IOException {
457        final HttpClient urlConn = HttpClient.create(getUrl(), requestMethod);
458        urlConn.setAccept("text/html, image/png, image/jpeg, image/gif, */*");
459        urlConn.setReadTimeout(readTimeout); // 30 seconds read timeout
460        urlConn.setConnectTimeout(connectTimeout);
461        if (headers != null) {
462            urlConn.setHeaders(headers);
463        }
464
465        if (force || noCache) {
466            urlConn.useCache(false);
467        }
468        return urlConn;
469    }
470
471    private boolean isCacheValidUsingHead() throws IOException {
472        final HttpClient.Response urlConn = getRequest("HEAD", false).connect();
473        long lastModified = urlConn.getLastModified();
474        return (attributes.getEtag() != null && attributes.getEtag().equals(urlConn.getHeaderField("ETag"))) ||
475                (lastModified != 0 && lastModified <= attributes.getLastModification());
476    }
477
478    /**
479     * TODO: move to JobFactory
480     * cancels all outstanding tasks in the queue.
481     */
482    public void cancelOutstandingTasks() {
483        for (Runnable r: downloadJobExecutor.getQueue()) {
484            if (downloadJobExecutor.remove(r) && r instanceof JCSCachedTileLoaderJob) {
485                ((JCSCachedTileLoaderJob<?, ?>) r).handleJobCancellation();
486            }
487        }
488    }
489
490    /**
491     * Sets a job, that will be run, when job will finish execution
492     * @param runnable that will be executed
493     */
494    public void setFinishedTask(Runnable runnable) {
495        this.finishTask = runnable;
496
497    }
498
499    /**
500     * Marks this job as canceled
501     */
502    public void handleJobCancellation() {
503        finishLoading(LoadResult.CANCELED);
504    }
505
506    private URL getUrlNoException() {
507        try {
508            return getUrl();
509        } catch (IOException e) {
510            return null;
511        }
512    }
513}