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.util.HashSet; 008import java.util.List; 009import java.util.Map; 010import java.util.Random; 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; 028 029/** 030 * @author Wiktor Niesiobędzki 031 * 032 * @param <K> cache entry key type 033 * @param <V> cache value type 034 * 035 * Generic loader for HTTP based tiles. Uses custom attribute, to check, if entry has expired 036 * according to HTTP headers sent with tile. If so, it tries to verify using Etags 037 * or If-Modified-Since / Last-Modified. 038 * 039 * If the tile is not valid, it will try to download it from remote service and put it 040 * to cache. If remote server will fail it will try to use stale entry. 041 * 042 * This class will keep only one Job running for specified tile. All others will just finish, but 043 * listeners will be gathered and notified, once download job will be finished 044 * 045 * @since 8168 046 */ 047public abstract class JCSCachedTileLoaderJob<K, V extends CacheEntry> implements ICachedLoaderJob<K>, Runnable { 048 private static final Logger log = FeatureAdapter.getLogger(JCSCachedTileLoaderJob.class.getCanonicalName()); 049 protected static final long DEFAULT_EXPIRE_TIME = 1000L * 60 * 60 * 24 * 7; // 7 days 050 // Limit for the max-age value send by the server. 051 protected static final long EXPIRE_TIME_SERVER_LIMIT = 1000L * 60 * 60 * 24 * 28; // 4 weeks 052 // Absolute expire time limit. Cached tiles that are older will not be used, 053 // even if the refresh from the server fails. 054 protected static final long ABSOLUTE_EXPIRE_TIME_LIMIT = 1000L * 60 * 60 * 24 * 365; // 1 year 055 056 /** 057 * maximum download threads that will be started 058 */ 059 public static final IntegerProperty THREAD_LIMIT = new IntegerProperty("cache.jcs.max_threads", 10); 060 061 /* 062 * ThreadPoolExecutor starts new threads, until THREAD_LIMIT is reached. Then it puts tasks into LinkedBlockingDeque. 063 * 064 * The queue works FIFO, so one needs to take care about ordering of the entries submitted 065 * 066 * 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 067 * the response, so later it could be used). We could actually cancel what is in LIFOQueue, but this is a tradeoff between simplicity 068 * and performance (we do want to have something to offer to worker threads before tasks will be resubmitted by class consumer) 069 */ 070 071 private static final ThreadPoolExecutor DEFAULT_DOWNLOAD_JOB_DISPATCHER = new ThreadPoolExecutor( 072 2, // we have a small queue, so threads will be quickly started (threads are started only, when queue is full) 073 THREAD_LIMIT.get(), // do not this number of threads 074 30, // keepalive for thread 075 TimeUnit.SECONDS, 076 // make queue of LIFO type - so recently requested tiles will be loaded first (assuming that these are which user is waiting to see) 077 new LinkedBlockingDeque<Runnable>(), 078 Utils.newThreadFactory("JCS-downloader-%d", Thread.NORM_PRIORITY) 079 ); 080 081 082 083 private static final ConcurrentMap<String, Set<ICachedLoaderListener>> inProgress = new ConcurrentHashMap<>(); 084 private static final ConcurrentMap<String, Boolean> useHead = new ConcurrentHashMap<>(); 085 086 protected final long now; // when the job started 087 088 private final ICacheAccess<K, V> cache; 089 private ICacheElement<K, V> cacheElement; 090 protected V cacheData; 091 protected CacheEntryAttributes attributes; 092 093 // HTTP connection parameters 094 private final int connectTimeout; 095 private final int readTimeout; 096 private final Map<String, String> headers; 097 private final ThreadPoolExecutor downloadJobExecutor; 098 private Runnable finishTask; 099 private boolean force; 100 101 /** 102 * @param cache cache instance that we will work on 103 * @param headers HTTP headers to be sent together with request 104 * @param readTimeout when connecting to remote resource 105 * @param connectTimeout when connecting to remote resource 106 * @param downloadJobExecutor that will be executing the jobs 107 */ 108 public JCSCachedTileLoaderJob(ICacheAccess<K, V> cache, 109 int connectTimeout, int readTimeout, 110 Map<String, String> headers, 111 ThreadPoolExecutor downloadJobExecutor) { 112 113 this.cache = cache; 114 this.now = System.currentTimeMillis(); 115 this.connectTimeout = connectTimeout; 116 this.readTimeout = readTimeout; 117 this.headers = headers; 118 this.downloadJobExecutor = downloadJobExecutor; 119 } 120 121 /** 122 * @param cache cache instance that we will work on 123 * @param headers HTTP headers to be sent together with request 124 * @param readTimeout when connecting to remote resource 125 * @param connectTimeout when connecting to remote resource 126 */ 127 public JCSCachedTileLoaderJob(ICacheAccess<K, V> cache, 128 int connectTimeout, int readTimeout, 129 Map<String, String> headers) { 130 this(cache, connectTimeout, readTimeout, 131 headers, DEFAULT_DOWNLOAD_JOB_DISPATCHER); 132 } 133 134 private void ensureCacheElement() { 135 if (cacheElement == null && getCacheKey() != null) { 136 cacheElement = cache.getCacheElement(getCacheKey()); 137 if (cacheElement != null) { 138 attributes = (CacheEntryAttributes) cacheElement.getElementAttributes(); 139 cacheData = cacheElement.getVal(); 140 } 141 } 142 } 143 144 public V get() { 145 ensureCacheElement(); 146 return cacheData; 147 } 148 149 @Override 150 public void submit(ICachedLoaderListener listener, boolean force) throws IOException { 151 this.force = force; 152 boolean first = false; 153 URL url = getUrl(); 154 String deduplicationKey = null; 155 if (url != null) { 156 // url might be null, for example when Bing Attribution is not loaded yet 157 deduplicationKey = url.toString(); 158 } 159 if (deduplicationKey == null) { 160 log.log(Level.WARNING, "No url returned for: {0}, skipping", getCacheKey()); 161 throw new IllegalArgumentException("No url returned"); 162 } 163 synchronized (inProgress) { 164 Set<ICachedLoaderListener> newListeners = inProgress.get(deduplicationKey); 165 if (newListeners == null) { 166 newListeners = new HashSet<>(); 167 inProgress.put(deduplicationKey, newListeners); 168 first = true; 169 } 170 newListeners.add(listener); 171 } 172 173 if (first || force) { 174 // submit all jobs to separate thread, so calling thread is not blocked with IO when loading from disk 175 log.log(Level.FINE, "JCS - Submitting job for execution for url: {0}", getUrlNoException()); 176 downloadJobExecutor.execute(this); 177 } 178 } 179 180 /** 181 * This method is run when job has finished 182 */ 183 protected void executionFinished() { 184 if (finishTask != null) { 185 finishTask.run(); 186 } 187 } 188 189 /** 190 * 191 * @return checks if object from cache has sufficient data to be returned 192 */ 193 protected boolean isObjectLoadable() { 194 if (cacheData == null) { 195 return false; 196 } 197 byte[] content = cacheData.getContent(); 198 return content != null && content.length > 0; 199 } 200 201 /** 202 * Simple implementation. All errors should be cached as empty. Though some JDK (JDK8 on Windows for example) 203 * doesn't return 4xx error codes, instead they do throw an FileNotFoundException or IOException 204 * 205 * @return true if we should put empty object into cache, regardless of what remote resource has returned 206 */ 207 protected boolean cacheAsEmpty() { 208 return attributes.getResponseCode() < 500; 209 } 210 211 /** 212 * @return key under which discovered server settings will be kept 213 */ 214 protected String getServerKey() { 215 return getUrlNoException().getHost(); 216 } 217 218 @Override 219 public void run() { 220 final Thread currentThread = Thread.currentThread(); 221 final String oldName = currentThread.getName(); 222 currentThread.setName("JCS Downloading: " + getUrlNoException()); 223 log.log(Level.FINE, "JCS - starting fetch of url: {0} ", getUrlNoException()); 224 ensureCacheElement(); 225 try { 226 // try to fetch from cache 227 if (!force && cacheElement != null && isCacheElementValid() && isObjectLoadable()) { 228 // we got something in cache, and it's valid, so lets return it 229 log.log(Level.FINE, "JCS - Returning object from cache: {0}", getCacheKey()); 230 finishLoading(LoadResult.SUCCESS); 231 return; 232 } 233 234 // try to load object from remote resource 235 if (loadObject()) { 236 finishLoading(LoadResult.SUCCESS); 237 } else { 238 // if loading failed - check if we can return stale entry 239 if (isObjectLoadable()) { 240 // try to get stale entry in cache 241 finishLoading(LoadResult.SUCCESS); 242 log.log(Level.FINE, "JCS - found stale object in cache: {0}", getUrlNoException()); 243 } else { 244 // failed completely 245 finishLoading(LoadResult.FAILURE); 246 } 247 } 248 } finally { 249 executionFinished(); 250 currentThread.setName(oldName); 251 } 252 } 253 254 private void finishLoading(LoadResult result) { 255 Set<ICachedLoaderListener> listeners = null; 256 synchronized (inProgress) { 257 listeners = inProgress.remove(getUrlNoException().toString()); 258 } 259 if (listeners == null) { 260 log.log(Level.WARNING, "Listener not found for URL: {0}. Listener not notified!", getUrlNoException()); 261 return; 262 } 263 for (ICachedLoaderListener l: listeners) { 264 l.loadingFinished(cacheData, attributes, result); 265 } 266 } 267 268 protected boolean isCacheElementValid() { 269 long expires = attributes.getExpirationTime(); 270 271 // check by expire date set by server 272 if (expires != 0L) { 273 // put a limit to the expire time (some servers send a value 274 // that is too large) 275 expires = Math.min(expires, attributes.getCreateTime() + EXPIRE_TIME_SERVER_LIMIT); 276 if (now > expires) { 277 log.log(Level.FINE, "JCS - Object {0} has expired -> valid to {1}, now is: {2}", 278 new Object[]{getUrlNoException(), Long.toString(expires), Long.toString(now)}); 279 return false; 280 } 281 } else if (attributes.getLastModification() > 0 && 282 now - attributes.getLastModification() > DEFAULT_EXPIRE_TIME) { 283 // check by file modification date 284 log.log(Level.FINE, "JCS - Object has expired, maximum file age reached {0}", getUrlNoException()); 285 return false; 286 } else if (now - attributes.getCreateTime() > DEFAULT_EXPIRE_TIME) { 287 log.log(Level.FINE, "JCS - Object has expired, maximum time since object creation reached {0}", getUrlNoException()); 288 return false; 289 } 290 return true; 291 } 292 293 /** 294 * @return true if object was successfully downloaded, false, if there was a loading failure 295 */ 296 297 private boolean loadObject() { 298 if (attributes == null) { 299 attributes = new CacheEntryAttributes(); 300 } 301 try { 302 // if we have object in cache, and host doesn't support If-Modified-Since nor If-None-Match 303 // then just use HEAD request and check returned values 304 if (isObjectLoadable() && 305 Boolean.TRUE.equals(useHead.get(getServerKey())) && 306 isCacheValidUsingHead()) { 307 log.log(Level.FINE, "JCS - cache entry verified using HEAD request: {0}", getUrl()); 308 return true; 309 } 310 311 final HttpClient request = getRequest("GET", true); 312 313 if (isObjectLoadable() && 314 (now - attributes.getLastModification()) <= ABSOLUTE_EXPIRE_TIME_LIMIT) { 315 request.setIfModifiedSince(attributes.getLastModification()); 316 } 317 if (isObjectLoadable() && attributes.getEtag() != null) { 318 request.setHeader("If-None-Match", attributes.getEtag()); 319 } 320 321 final HttpClient.Response urlConn = request.connect(); 322 323 if (urlConn.getResponseCode() == 304) { 324 // If isModifiedSince or If-None-Match has been set 325 // and the server answers with a HTTP 304 = "Not Modified" 326 log.log(Level.FINE, "JCS - If-Modified-Since/ETag test: local version is up to date: {0}", getUrl()); 327 return true; 328 } else if (isObjectLoadable() // we have an object in cache, but we haven't received 304 response code 329 && ( 330 (attributes.getEtag() != null && attributes.getEtag().equals(urlConn.getHeaderField("ETag"))) || 331 attributes.getLastModification() == urlConn.getLastModified()) 332 ) { 333 // we sent ETag or If-Modified-Since, but didn't get 304 response code 334 // for further requests - use HEAD 335 String serverKey = getServerKey(); 336 log.log(Level.INFO, "JCS - Host: {0} found not to return 304 codes for If-Modified-Since or If-None-Match headers", 337 serverKey); 338 useHead.put(serverKey, Boolean.TRUE); 339 } 340 341 342 attributes = parseHeaders(urlConn); 343 344 for (int i = 0; i < 5; ++i) { 345 if (urlConn.getResponseCode() == 503) { 346 Thread.sleep(5000+(new Random()).nextInt(5000)); 347 continue; 348 } 349 350 attributes.setResponseCode(urlConn.getResponseCode()); 351 byte[] raw; 352 if (urlConn.getResponseCode() == 200) { 353 raw = Utils.readBytesFromStream(urlConn.getContent()); 354 } else { 355 raw = new byte[]{}; 356 } 357 358 if (isResponseLoadable(urlConn.getHeaderFields(), urlConn.getResponseCode(), raw)) { 359 // we need to check cacheEmpty, so for cases, when data is returned, but we want to store 360 // as empty (eg. empty tile images) to save some space 361 cacheData = createCacheEntry(raw); 362 cache.put(getCacheKey(), cacheData, attributes); 363 log.log(Level.FINE, "JCS - downloaded key: {0}, length: {1}, url: {2}", 364 new Object[] {getCacheKey(), raw.length, getUrl()}); 365 return true; 366 } else if (cacheAsEmpty()) { 367 cacheData = createCacheEntry(new byte[]{}); 368 cache.put(getCacheKey(), cacheData, attributes); 369 log.log(Level.FINE, "JCS - Caching empty object {0}", getUrl()); 370 return true; 371 } else { 372 log.log(Level.FINE, "JCS - failure during load - reponse is not loadable nor cached as empty"); 373 return false; 374 } 375 } 376 } catch (FileNotFoundException e) { 377 log.log(Level.FINE, "JCS - Caching empty object as server returned 404 for: {0}", getUrlNoException()); 378 attributes.setResponseCode(404); 379 attributes.setErrorMessage(e.toString()); 380 boolean doCache = isResponseLoadable(null, 404, null) || cacheAsEmpty(); 381 if (doCache) { 382 cacheData = createCacheEntry(new byte[]{}); 383 cache.put(getCacheKey(), cacheData, attributes); 384 } 385 return doCache; 386 } catch (IOException e) { 387 log.log(Level.FINE, "JCS - IOExecption during communication with server for: {0}", getUrlNoException()); 388 attributes.setErrorMessage(e.toString()); 389 attributes.setResponseCode(499); // set dummy error code 390 boolean doCache = isResponseLoadable(null, 499, null) || cacheAsEmpty(); //generic 499 error code returned 391 if (doCache) { 392 cacheData = createCacheEntry(new byte[]{}); 393 cache.put(getCacheKey(), createCacheEntry(new byte[]{}), attributes); 394 } 395 return doCache; 396 } catch (Exception e) { 397 attributes.setErrorMessage(e.toString()); 398 log.log(Level.WARNING, "JCS - Exception during download {0}", getUrlNoException()); 399 Main.warn(e); 400 } 401 log.log(Level.WARNING, "JCS - Silent failure during download: {0}", getUrlNoException()); 402 return false; 403 404 } 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}