log4j2 自动删除过期日志文件配置及实现原理解析

　　日志文件自动删除功能必不可少，当然你可以让运维去做这事，只是这不地道。而日志组件是一个必备组件，让其多做一件删除的工作，无可厚非。本文就来探讨下 log4j 的日志文件自动删除实现吧。

0. 自动删除配置参考样例: （log4j2.xml）

<?xml version="1.0" encoding="UTF-8" ?>
<Configuration status="warn" monitorInterval="30" strict="true"
               schema="Log4J-V2.2.xsd">
    <Properties>
        <Property name="log_level">info</Property>
    </Properties>
    <Appenders>
        <!-- 输出到控制台 -->
        <Console name="Console" target="SYSTEM_OUT">
            <ThresholdFilter level="${log_level}" onMatch="ACCEPT" onMismatch="DENY" />
            <PatternLayout pattern="%d{yyyy-MM-dd'T'HH:mm:ss.SSS} [%t] %p - %c - %m%n" />
        </Console>
        <!-- 与properties文件中位置存在冲突，如有问题，请注意调整 -->
        <RollingFile name="logFile" fileName="logs/app/test.log"
                     filePattern="logs/app/history/test-%d{MM-dd-yyyy}-%i.log.gz">
            <ThresholdFilter level="${log_level}" onMatch="ACCEPT" onMismatch="DENY"  />
            <PatternLayout pattern="%d{yyyy-MM-dd'T'HH:mm:ss.SSS} [%p] [%c:%L] -- %m%n" />
            <Policies>
                <!-- 按天递计算频率 -->
                <TimeBasedTriggeringPolicy interval="1" />
                <SizeBasedTriggeringPolicy size="500 MB" />
                <OnStartupTriggeringPolicy />
            </Policies>
            <!-- 删除策略配置 -->
            <DefaultRolloverStrategy max="5">
                <Delete basePath="logs/app/history" maxDepth="1">
                    <IfFileName glob="*.log.gz"/>
                    <IfLastModified age="7d"/>
                </Delete>
                <Delete basePath="logs/app/history" maxDepth="1">
                    <IfFileName glob="*.docx"/>
                </Delete>
                <Delete basePath="logs/app/history" maxDepth="1">
                    <IfFileName glob="*.vsdx"/>
                </Delete>
            </DefaultRolloverStrategy>
        </RollingFile>
        <Async name="Async" bufferSize="2000" blocking="false">
            <AppenderRef ref="logFile"/>
        </Async>
    </Appenders>

    <Loggers>
        <Root level="${log_level}">
            <AppenderRef ref="Console" />
            <AppenderRef ref="Async" />
        </Root>
        <!-- 配置个例 -->
        <Logger name="com.xx.filter" level="info" />
    </Loggers>
</Configuration>

　　如果仅想停留在使用层面，如上log4j2.xml配置文件足矣！

　　不过，至少得注意一点，以上配置需要基于log4j2, 而如果你是 log4j1.x,则需要做下无缝升级：主要就是换下jar包版本，换个桥接包之类的,比如下参考配置：

    
            <dependency>
                <groupId>commons-logging</groupId>
                <artifactId>commons-logging</artifactId>
                <version>1.2</version>
            </dependency>
            <!-- 桥接：告诉commons logging使用Log4j2 -->
            <dependency>
                <groupId>org.slf4j</groupId>
                <artifactId>jcl-over-slf4j</artifactId>
                <version>1.7.26</version>
            </dependency>

            <!-- 此处老版本，需注释掉 -->
            <!--<dependency>
                <groupId>log4j</groupId>
                <artifactId>log4j</artifactId>
                <version>1.2.17</version>
            </dependency>-->

            <dependency>
                <groupId>org.apache.commons</groupId>
                <artifactId>commons-compress</artifactId>
                <version>1.10</version>
            </dependency>
            <dependency>
                <groupId>org.apache.logging.log4j</groupId>
                <artifactId>log4j-core</artifactId>
                <version>2.8.2</version>
            </dependency>
            <dependency>
                <groupId>org.apache.logging.log4j</groupId>
                <artifactId>log4j-slf4j-impl</artifactId>
                <version>2.8.2</version>
            </dependency>
            <dependency>
                <groupId>org.apache.logging.log4j</groupId>
                <artifactId>log4j-api</artifactId>
                <version>2.8.2</version>
            </dependency>
            <dependency>
                <groupId>org.apache.logging.log4j</groupId>
                <artifactId>log4j-web</artifactId>
                <version>2.8.2</version>
            </dependency>

　　如果还想多了解一点其运行原理，就跟随本文的脚步吧:

1. 自动清理大体运行流程

　　自动删除工作的运行原理大体流程如下。（大抵都是如此）

　　　　1. 加载log4j2.xml配置文件;
　　　　2. 读取appenders，并添加到log4j上下文中;
　　　　3. 加载 policy, 加载 rollover 配置;
　　　　4. 写入日志时判断是否满足rollover配置, 默认是一天运行一次, 可自行添加各种运行测试, 比如大小、启动时;

　　所以，删除策略的核心是每一次添加日志时。代码验证如下：

    // 在每次添加日志时判定
    // org.apache.logging.log4j.core.appender.RollingRandomAccessFileAppender#append
    /**
     * Write the log entry rolling over the file when required.
     *
     * @param event The LogEvent.
     */
    @Override
    public void append(final LogEvent event) {
        final RollingRandomAccessFileManager manager = getManager();
        // 重点：直接检查是否需要 rollover, 如需要直接进行
        manager.checkRollover(event);

        // Leverage the nice batching behaviour of async Loggers/Appenders:
        // we can signal the file manager that it needs to flush the buffer
        // to disk at the end of a batch.
        // From a user's point of view, this means that all log events are
        // _always_ available in the log file, without incurring the overhead
        // of immediateFlush=true.
        manager.setEndOfBatch(event.isEndOfBatch()); // FIXME manager's EndOfBatch threadlocal can be deleted

        // LOG4J2-1292 utilize gc-free Layout.encode() method: taken care of in superclass
        super.append(event);
    }

    // org.apache.logging.log4j.core.appender.rolling.RollingFileManager#checkRollover
    /**
     * Determines if a rollover should occur.
     * @param event The LogEvent.
     */
    public synchronized void checkRollover(final LogEvent event) {
        // 由各触发策略判定是否需要进行 rolling
        // 如需要, 则调用 rollover()
        if (triggeringPolicy.isTriggeringEvent(event)) {
            rollover();
        }
    }

　　所以，何时进行删除？答案是在适当的时机，这个时机可以是任意时候。

2. log4j 日志滚动

　　日志滚动，可以是重命名，也可以是删除文件。但总体判断是否可触发滚动的前提是一致的。我们这里主要关注文件删除。我们以时间作为依据看下判断过程。

    // 1. 判断是否是 触发事件时机
    // org.apache.logging.log4j.core.appender.rolling.TimeBasedTriggeringPolicy#isTriggeringEvent
    /**
     * Determines whether a rollover should occur.
     * @param event   A reference to the currently event.
     * @return true if a rollover should occur.
     */
    @Override
    public boolean isTriggeringEvent(final LogEvent event) {
        if (manager.getFileSize() == 0) {
            return false;
        }
        final long nowMillis = event.getTimeMillis();
        // TimeBasedTriggeringPolicy, 是基于时间判断的, 此处为每天一次
        if (nowMillis >= nextRolloverMillis) {
            nextRolloverMillis = manager.getPatternProcessor().getNextTime(nowMillis, interval, modulate);
            return true;
        }
        return false;
    }
    // org.apache.logging.log4j.core.appender.rolling.RollingFileManager#rollover()
    public synchronized void rollover() {
        if (!hasOutputStream()) {
            return;
        }
        // strategy 是xml配置的策略
        if (rollover(rolloverStrategy)) {
            try {
                size = 0;
                initialTime = System.currentTimeMillis();
                createFileAfterRollover();
            } catch (final IOException e) {
                logError("Failed to create file after rollover", e);
            }
        }
    }
    // RollingFileManager 统一管理触发器
    // org.apache.logging.log4j.core.appender.rolling.RollingFileManager#rollover
    private boolean rollover(final RolloverStrategy strategy) {

        boolean releaseRequired = false;
        try {
            // Block until the asynchronous operation is completed.
            // 上锁保证线程安全
            semaphore.acquire();
            releaseRequired = true;
        } catch (final InterruptedException e) {
            logError("Thread interrupted while attempting to check rollover", e);
            return false;
        }

        boolean success = true;

        try {
            // 由各触发器运行 rollover 逻辑
            final RolloverDescription descriptor = strategy.rollover(this);
            if (descriptor != null) {
                writeFooter();
                closeOutputStream();
                if (descriptor.getSynchronous() != null) {
                    LOGGER.debug("RollingFileManager executing synchronous {}", descriptor.getSynchronous());
                    try {
                        // 先使用同步方法，改名，然后再使用异步方法操作更多
                        success = descriptor.getSynchronous().execute();
                    } catch (final Exception ex) {
                        success = false;
                        logError("Caught error in synchronous task", ex);
                    }
                }
                // 如果配置了异步器, 则使用异步进行 rollover
                if (success && descriptor.getAsynchronous() != null) {
                    LOGGER.debug("RollingFileManager executing async {}", descriptor.getAsynchronous());
                    // CompositeAction, 使用异步线程池运行用户的 action
                    asyncExecutor.execute(new AsyncAction(descriptor.getAsynchronous(), this));
                    // 在异步运行action期间，锁是不会被释放的，以避免线程安全问题
                    // 直到异步任务完成，再主动释放锁
                    releaseRequired = false;
                }
                return true;
            }
            return false;
        } finally {
            if (releaseRequired) {
                semaphore.release();
            }
        }

    }

　　此处滚动有两个处理点，1. 每个滚动策略可以自行处理业务； 2. RollingFileManager 统一管理触发同步和异步的滚动action；

3. DefaultRolloverStrategy 默认滚动策略驱动

　　DefaultRolloverStrategy 作为一个默认的滚动策略实现，可以配置多个 Action, 然后处理删除操作。

　　删除有两种方式: 1. 当次滚动的文件数过多，会立即进行删除; 2. 配置单独的 DeleteAction, 根据配置的具体策略进行删除。（但该Action只会被返回给外部调用，自身则不会执行）

    // org.apache.logging.log4j.core.appender.rolling.DefaultRolloverStrategy#rollover
    /**
     * Performs the rollover.
     *
     * @param manager The RollingFileManager name for current active log file.
     * @return A RolloverDescription.
     * @throws SecurityException if an error occurs.
     */
    @Override
    public RolloverDescription rollover(final RollingFileManager manager) throws SecurityException {
        int fileIndex;
        // 默认 minIndex=1
        if (minIndex == Integer.MIN_VALUE) {
            final SortedMap<Integer, Path> eligibleFiles = getEligibleFiles(manager);
            fileIndex = eligibleFiles.size() > 0 ? eligibleFiles.lastKey() + 1 : 1;
        } else {
            if (maxIndex < 0) {
                return null;
            }
            final long startNanos = System.nanoTime();
            // 删除case1: 获取符合条件的文件数，同时清理掉大于  max 配置的日志文件
            // 如配置 max=5, 当前只有4个满足时, 不会立即清理文件, 但也不会阻塞后续流程
            // 只要没有出现错误, fileIndex 不会小于0
            fileIndex = purge(minIndex, maxIndex, manager);
            if (fileIndex < 0) {
                return null;
            }
            if (LOGGER.isTraceEnabled()) {
                final double durationMillis = TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - startNanos);
                LOGGER.trace("DefaultRolloverStrategy.purge() took {} milliseconds", durationMillis);
            }
        }
        // 进入此区域即意味着，必然有文件需要滚动，重新命名了
        final StringBuilder buf = new StringBuilder(255);
        manager.getPatternProcessor().formatFileName(strSubstitutor, buf, fileIndex);
        final String currentFileName = manager.getFileName();

        String renameTo = buf.toString();
        final String compressedName = renameTo;
        Action compressAction = null;

        FileExtension fileExtension = manager.getFileExtension();
        if (fileExtension != null) {
            renameTo = renameTo.substring(0, renameTo.length() - fileExtension.length());
            compressAction = fileExtension.createCompressAction(renameTo, compressedName,
                    true, compressionLevel);
        }
        // 未发生文件重命名情况，即文件未被重命名未被滚动
        // 该种情况应该不太会发生
        if (currentFileName.equals(renameTo)) {
            LOGGER.warn("Attempt to rename file {} to itself will be ignored", currentFileName);
            return new RolloverDescriptionImpl(currentFileName, false, null, null);
        }
        // 新建一个重命令的 action, 返回待用
        final FileRenameAction renameAction = new FileRenameAction(new File(currentFileName), new File(renameTo),
                    manager.isRenameEmptyFiles());
        // 异步处理器，会处理用户配置的异步action，如本文配置的 DeleteAction
        // 它将会在稍后被提交到异步线程池中运行
        final Action asyncAction = merge(compressAction, customActions, stopCustomActionsOnError);
        // 封装Rollover返回, renameAction 是同步方法, 其他用户配置的动态action 则是异步方法
        // 删除case2: 封装异步返回action
        return new RolloverDescriptionImpl(currentFileName, false, renameAction, asyncAction);
    }
    private int purge(final int lowIndex, final int highIndex, final RollingFileManager manager) {
        // 默认使用 accending 的方式进行清理文件
        return useMax ? purgeAscending(lowIndex, highIndex, manager) : purgeDescending(lowIndex, highIndex, manager);
    }
    // org.apache.logging.log4j.core.appender.rolling.DefaultRolloverStrategy#purgeAscending
    /**
     * Purges and renames old log files in preparation for rollover. The oldest file will have the smallest index, the
     * newest the highest.
     *
     * @param lowIndex low index. Log file associated with low index will be deleted if needed.
     * @param highIndex high index.
     * @param manager The RollingFileManager
     * @return true if purge was successful and rollover should be attempted.
     */
    private int purgeAscending(final int lowIndex, final int highIndex, final RollingFileManager manager) {
        final SortedMap<Integer, Path> eligibleFiles = getEligibleFiles(manager);
        final int maxFiles = highIndex - lowIndex + 1;

        boolean renameFiles = false;
        // 依次迭代 eligibleFiles, 删除
        while (eligibleFiles.size() >= maxFiles) {
            try {
                LOGGER.debug("Eligible files: {}", eligibleFiles);
                Integer key = eligibleFiles.firstKey();
                LOGGER.debug("Deleting {}", eligibleFiles.get(key).toFile().getAbsolutePath());
                // 调用nio的接口删除文件
                Files.delete(eligibleFiles.get(key));
                eligibleFiles.remove(key);
                renameFiles = true;
            } catch (IOException ioe) {
                LOGGER.error("Unable to delete {}, {}", eligibleFiles.firstKey(), ioe.getMessage(), ioe);
                break;
            }
        }
        final StringBuilder buf = new StringBuilder();
        if (renameFiles) {
            // 针对未完成删除的文件，继续处理
            // 比如使用 匹配的方式匹配文件, 则不能被正常删除
            // 还有些未超过maxFiles的文件
            for (Map.Entry<Integer, Path> entry : eligibleFiles.entrySet()) {
                buf.setLength(0);
                // LOG4J2-531: directory scan & rollover must use same format
                manager.getPatternProcessor().formatFileName(strSubstitutor, buf, entry.getKey() - 1);
                String currentName = entry.getValue().toFile().getName();
                String renameTo = buf.toString();
                int suffixLength = suffixLength(renameTo);
                if (suffixLength > 0 && suffixLength(currentName) == 0) {
                   renameTo = renameTo.substring(0, renameTo.length() - suffixLength);
                }
                Action action = new FileRenameAction(entry.getValue().toFile(), new File(renameTo), true);
                try {
                    LOGGER.debug("DefaultRolloverStrategy.purgeAscending executing {}", action);
                    if (!action.execute()) {
                        return -1;
                    }
                } catch (final Exception ex) {
                    LOGGER.warn("Exception during purge in RollingFileAppender", ex);
                    return -1;
                }
            }
        }
        // 此处返回的 findIndex 一定是 >=0 的
        return eligibleFiles.size() > 0 ?
                (eligibleFiles.lastKey() < highIndex ? eligibleFiles.lastKey() + 1 : highIndex) : lowIndex;
    }

4. 符合过滤条件的文件查找

　　当配置了 max 参数，这个参数是如何匹配的呢？比如我某个文件夹下有很历史文件，是否都会匹配呢？

    // 文件查找规则 
    // org.apache.logging.log4j.core.appender.rolling.AbstractRolloverStrategy#getEligibleFiles
    protected SortedMap<Integer, Path> getEligibleFiles(final RollingFileManager manager) {
        return getEligibleFiles(manager, true);
    }
    protected SortedMap<Integer, Path> getEligibleFiles(final RollingFileManager manager,
                                                        final boolean isAscending) {
        final StringBuilder buf = new StringBuilder();
        // 此处的pattern 即是在appender上配置的 filePattern, 一般会受限于 MM-dd-yyyy-$i.log.gz
        String pattern = manager.getPatternProcessor().getPattern();
        // 此处会将时间替换为当前, 然后按照此规则进行匹配要处理的文件
        manager.getPatternProcessor().formatFileName(strSubstitutor, buf, NotANumber.NAN);
        return getEligibleFiles(buf.toString(), pattern, isAscending);
    }
    // 细节匹配要处理的文件
    protected SortedMap<Integer, Path> getEligibleFiles(String path, String logfilePattern, boolean isAscending) {
        TreeMap<Integer, Path> eligibleFiles = new TreeMap<>();
        File file = new File(path);
        File parent = file.getParentFile();
        if (parent == null) {
            parent = new File(".");
        } else {
            parent.mkdirs();
        }
        if (!logfilePattern.contains("%i")) {
            return eligibleFiles;
        }
        Path dir = parent.toPath();
        String fileName = file.getName();
        int suffixLength = suffixLength(fileName);
        if (suffixLength > 0) {
            fileName = fileName.substring(0, fileName.length() - suffixLength) + ".*";
        }
        String filePattern = fileName.replace(NotANumber.VALUE, "(\d+)");
        Pattern pattern = Pattern.compile(filePattern);

        try (DirectoryStream<Path> stream = Files.newDirectoryStream(dir)) {
            for (Path entry: stream) {
                // 该匹配相当精确
                // 只会删除当天或者在时间交替的时候删除上一天的数据咯
                // 如果在这个时候进行了重启操作，就再也不会删除此文件了
                Matcher matcher = pattern.matcher(entry.toFile().getName());
                if (matcher.matches()) {
                    Integer index = Integer.parseInt(matcher.group(1));
                    eligibleFiles.put(index, entry);
                }
            }
        } catch (IOException ioe) {
            throw new LoggingException("Error reading folder " + dir + " " + ioe.getMessage(), ioe);
        }
        return isAscending? eligibleFiles : eligibleFiles.descendingMap();
    }
    // 此处会将 各种格式的文件名，替换为当前时间或者最后一次滚动的文件的时间。所以匹配的时候，并不会匹配超时当前认知范围的文件
    /**
     * Formats file name.
     * @param subst The StrSubstitutor.
     * @param buf string buffer to which formatted file name is appended, may not be null.
     * @param obj object to be evaluated in formatting, may not be null.
     */
    public final void formatFileName(final StrSubstitutor subst, final StringBuilder buf, final boolean useCurrentTime,
                                     final Object obj) {
        // LOG4J2-628: we deliberately use System time, not the log4j.Clock time
        // for creating the file name of rolled-over files.
        final long time = useCurrentTime && currentFileTime != 0 ? currentFileTime :
                prevFileTime != 0 ? prevFileTime : System.currentTimeMillis();
        formatFileName(buf, new Date(time), obj);
        final LogEvent event = new Log4jLogEvent.Builder().setTimeMillis(time).build();
        final String fileName = subst.replace(event, buf);
        buf.setLength(0);
        buf.append(fileName);
    }
    

　　AsyncAction 是一个 Runnable 的实现, 被直接提交到线程池运行. AsyncAction -> AbstractAction -> Action -> Runnable

　　它是一个统一管理异步Action的包装，主要是管理锁和异常类操作。

    // org.apache.logging.log4j.core.appender.rolling.RollingFileManager.AsyncAction
    /**
     * Performs actions asynchronously.
     */
    private static class AsyncAction extends AbstractAction {

        private final Action action;
        private final RollingFileManager manager;

        /**
         * Constructor.
         * @param act The action to perform.
         * @param manager The manager.
         */
        public AsyncAction(final Action act, final RollingFileManager manager) {
            this.action = act;
            this.manager = manager;
        }

        /**
         * Executes an action.
         *
         * @return true if action was successful.  A return value of false will cause
         *         the rollover to be aborted if possible.
         * @throws java.io.IOException if IO error, a thrown exception will cause the rollover
         *                             to be aborted if possible.
         */
        @Override
        public boolean execute() throws IOException {
            try {
                // 门面调用 action.execute(), 一般是调用  CompositeAction, 里面封装了多个 action
                return action.execute();
            } finally {
                // 任务执行完成，才会释放外部的锁
                // 虽然不是很优雅，但是很准确很安全
                manager.semaphore.release();
            }
        }
        ...
    }

    // CompositeAction 封装了多个 action 处理
    // org.apache.logging.log4j.core.appender.rolling.action.CompositeAction#run
    /**
     * Execute sequence of actions.
     *
     * @return true if all actions were successful.
     * @throws IOException on IO error.
     */
    @Override
    public boolean execute() throws IOException {
        if (stopOnError) {
            // 依次调用action
            for (final Action action : actions) {
                if (!action.execute()) {
                    return false;
                }
            }

            return true;
        }
        boolean status = true;
        IOException exception = null;

        for (final Action action : actions) {
            try {
                status &= action.execute();
            } catch (final IOException ex) {
                status = false;

                if (exception == null) {
                    exception = ex;
                }
            }
        }

        if (exception != null) {
            throw exception;
        }

        return status;
    }

　　DeleteAction 是我们真正关心的动作。

    // CompositeAction 封装了多个 action 处理
    // org.apache.logging.log4j.core.appender.rolling.action.CompositeAction#run
    /**
     * Execute sequence of actions.
     *
     * @return true if all actions were successful.
     * @throws IOException on IO error.
     */
    @Override
    public boolean execute() throws IOException {
        if (stopOnError) {
            // 依次调用action
            for (final Action action : actions) {
                if (!action.execute()) {
                    return false;
                }
            }

            return true;
        }
        boolean status = true;
        IOException exception = null;

        for (final Action action : actions) {
            try {
                status &= action.execute();
            } catch (final IOException ex) {
                status = false;

                if (exception == null) {
                    exception = ex;
                }
            }
        }

        if (exception != null) {
            throw exception;
        }

        return status;
    }
    
    // DeleteAction 做真正的删除动作
    // org.apache.logging.log4j.core.appender.rolling.action.DeleteAction#execute()
    @Override
    public boolean execute() throws IOException {
        // 如果没有script配置，则直接委托父类处理
        return scriptCondition != null ? executeScript() : super.execute();
    }
    org.apache.logging.log4j.core.appender.rolling.action.AbstractPathAction#execute()
    @Override
    public boolean execute() throws IOException {
        // 根据指定的basePath, 和过滤条件，选择相关文件
        // 调用 DeleteAction 的 createFileVisitor(), 返回 DeletingVisitor
        return execute(createFileVisitor(getBasePath(), pathConditions));
    }
    // org.apache.logging.log4j.core.appender.rolling.action.DeleteAction#execute(java.nio.file.FileVisitor<java.nio.file.Path>)
    @Override
    public boolean execute(final FileVisitor<Path> visitor) throws IOException {
        // 根据maxDepth设置，遍历所有可能的文件路径
        // 使用 Files.walkFileTree() 实现, 添加到 collected 中
        final List<PathWithAttributes> sortedPaths = getSortedPaths();
        trace("Sorted paths:", sortedPaths);

        for (final PathWithAttributes element : sortedPaths) {
            try {
                // 依次调用 visitFile, 依次判断是否需要删除
                visitor.visitFile(element.getPath(), element.getAttributes());
            } catch (final IOException ioex) {
                LOGGER.error("Error in post-rollover Delete when visiting {}", element.getPath(), ioex);
                visitor.visitFileFailed(element.getPath(), ioex);
            }
        }
        // TODO return (visitor.success || ignoreProcessingFailure)
        return true; // do not abort rollover even if processing failed
    }

　　最终，即和想像的一样：找到要查找的文件夹，遍历各文件，用多个条件判断是否满足。删除符合条件的文件。

　　只是这其中注意的点：如何删除文件的线程安全性；如何保证删除工作不影响业务线程；很常见的锁和多线程的应用。

　　

5. 真正的删除

　　真正的删除动作就是在 DeleteAction中配置的，但上面可以看它是调用 visitor 的visitFile 方法，所以有必要看看是如何真正处理删除的。（实际上前面在 purge时已经做过一次删除操作了，所以别被两个点迷惑了，建议尽量只依赖于 Delete 配置，可以将外部 max 设置很大以避免两处生效）

    // org.apache.logging.log4j.core.appender.rolling.action.DeletingVisitor#visitFile
    @Override
    public FileVisitResult visitFile(final Path file, final BasicFileAttributes attrs) throws IOException {
        for (final PathCondition pathFilter : pathConditions) {
            final Path relative = basePath.relativize(file);
            // 遍历所有条件，只要有一个不符合，即不进行删除。
            // 所以，所以条件是 AND 关系， 没有 OR 关系
            // 如果想配置 OR 关系，只能配置多个DELETE
            if (!pathFilter.accept(basePath, relative, attrs)) {
                LOGGER.trace("Not deleting base={}, relative={}", basePath, relative);
                return FileVisitResult.CONTINUE;
            }
        }
        // 直接删除文件
        if (isTestMode()) {
            LOGGER.info("Deleting {} (TEST MODE: file not actually deleted)", file);
        } else {
            delete(file);
        }
        return FileVisitResult.CONTINUE;
    }

　　删除策略配置比如:

    <RollingFile name="logFile" fileName="logs/app/test.log"
                 filePattern="logs/app/history/test-%d{MM-dd-yyyy}-%i.log.gz">
        <ThresholdFilter level="${log_level}" onMatch="ACCEPT" onMismatch="DENY"  />
        <PatternLayout pattern="%d{yyyy-MM-dd'T'HH:mm:ss.SSS} [%p] [%c:%L] -- %m%n" />
        <Policies>
            <!-- 按天递计算频率 -->
            <TimeBasedTriggeringPolicy interval="1" />
            <SizeBasedTriggeringPolicy size="500 MB" />
            <OnStartupTriggeringPolicy />
        </Policies>
        <!-- 删除策略配置 -->
        <DefaultRolloverStrategy max="5000">
            <Delete basePath="logs/app/history" maxDepth="1">
                <!-- 配置且关系 -->
                <IfFileName glob="*.log.gz"/>
                <IfLastModified age="7d"/>
            </Delete>
            <!-- 配置或关系 -->
            <Delete basePath="logs/app/history" maxDepth="1">
                <IfFileName glob="*.docx"/>
            </Delete>
            <Delete basePath="logs/app/history" maxDepth="1">
                <IfFileName glob="*.vsdx"/>
            </Delete>
        </DefaultRolloverStrategy>
    </RollingFile>

　　另外说明，日志实现之所以能够无缝升级或替换，是因为利用了不同实现版本的 org/slf4j/impl/StaticLoggerBinder.class, 而外部都使用 slf4j 接口定义实现的，比如 org.apache.logging.log4j:log4j-slf4j-impl 包的实现。这是因为 org.slf4j.LoggerFactory 中留了这么个坑位，等你来填的好处。

      我之前也有另外一篇文章介绍logback如何进行日志滚动删除的，有兴趣可以了解一下： https://www.cnblogs.com/yougewe/p/10343855.html