Examples with CorruptedDataStructureException org.apache.ignite.internal.processors.cache.persistence.CorruptedDataStructureException used on opensource projects

Example 1 with CorruptedDataStructureException

use of org.apache.ignite.internal.processors.cache.persistence.CorruptedDataStructureException in project ignite by apache.

the class DiagnosticProcessor method onFailure.

/**
 * Print diagnostic info about failure occurred on {@code ignite} instance.
 * Failure details is contained in {@code failureCtx}.
 *
 * @param failureCtx Failure context.
 */
public void onFailure(FailureContext failureCtx) {
    // Dump data structures page locks.
    if (IGNITE_DUMP_PAGE_LOCK_ON_FAILURE)
        ctx.cache().context().diagnostic().pageLockTracker().dumpLocksToLog();
    CorruptedDataStructureException corruptedDataStructureEx = X.cause(failureCtx.error(), CorruptedDataStructureException.class);
    if (corruptedDataStructureEx != null && !F.isEmpty(corruptedDataStructureEx.pageIds()) && fileIOFactory != null) {
        File[] walDirs = walDirs(ctx);
        if (F.isEmpty(walDirs)) {
            if (log.isInfoEnabled())
                log.info("Skipping dump diagnostic info due to WAL not configured");
        } else {
            try {
                File corruptedPagesFile = corruptedPagesFile(diagnosticPath, fileIOFactory, corruptedDataStructureEx.groupId(), corruptedDataStructureEx.pageIds());
                String walDirsStr = Arrays.stream(walDirs).map(File::getAbsolutePath).collect(joining(", ", "[", "]"));
                String args = "walDir=" + walDirs[0].getAbsolutePath() + (walDirs.length == 1 ? "" : " walArchiveDir=" + walDirs[1].getAbsolutePath());
                if (ctx.config().getDataStorageConfiguration().getPageSize() != DFLT_PAGE_SIZE)
                    args += " pageSize=" + ctx.config().getDataStorageConfiguration().getPageSize();
                args += " pages=" + corruptedPagesFile.getAbsolutePath();
                log.warning(corruptedDataStructureEx.getClass().getSimpleName() + " has occurred. " + "To diagnose it, make a backup of the following directories: " + walDirsStr + ". " + "Then, run the following command: java -cp <classpath> " + "org.apache.ignite.development.utils.IgniteWalConverter " + args);
            } catch (Throwable t) {
                String pages = LongStream.of(corruptedDataStructureEx.pageIds()).mapToObj(pageId -> corruptedDataStructureEx.groupId() + ":" + pageId).collect(joining("\n", "", ""));
                log.error("Failed to dump diagnostic info of partition corruption. Page ids:\n" + pages, t);
            }
        }
    }
}

Example 2 with CorruptedDataStructureException

use of org.apache.ignite.internal.processors.cache.persistence.CorruptedDataStructureException in project ignite by apache.

the class FailureProcessor method process.

/**
 * Processes failure accordingly to given failure handler.
 *
 * @param failureCtx Failure context.
 * @param hnd Failure handler.
 * @return {@code True} If this very call led to Ignite node invalidation.
 */
public synchronized boolean process(FailureContext failureCtx, FailureHandler hnd) {
    assert failureCtx != null;
    assert hnd != null;
    if (// Node already terminating, no reason to process more errors.
    this.failureCtx != null)
        return false;
    if (failureTypeIgnored(failureCtx, hnd)) {
        U.quietAndWarn(ignite.log(), IGNORED_FAILURE_LOG_MSG + "[hnd=" + hnd + ", failureCtx=" + failureCtx + ']', failureCtx.error());
    } else {
        U.error(ignite.log(), FAILURE_LOG_MSG + "[hnd=" + hnd + ", failureCtx=" + failureCtx + ']', failureCtx.error());
    }
    if (reserveBuf != null && X.hasCause(failureCtx.error(), OutOfMemoryError.class))
        reserveBuf = null;
    CorruptedDataStructureException corruptedDataStructureEx = X.cause(failureCtx.error(), CorruptedDataStructureException.class);
    if (corruptedDataStructureEx != null) {
        CacheGroupContext grpCtx = ctx.cache().cacheGroup(corruptedDataStructureEx.groupId());
        if (grpCtx != null && grpCtx.dataRegion() != null) {
            if (grpCtx.dataRegion().config().isPersistenceEnabled()) {
                log.error("A critical problem with persistence data structures was detected." + " Please make backup of persistence storage and WAL files for further analysis." + " Persistence storage path: " + ctx.config().getDataStorageConfiguration().getStoragePath() + " WAL path: " + ctx.config().getDataStorageConfiguration().getWalPath() + " WAL archive path: " + ctx.config().getDataStorageConfiguration().getWalArchivePath());
            } else
                log.error("A critical problem with in-memory data structures was detected.");
        }
    }
    if (igniteDumpThreadsOnFailure && !throttleThreadDump(failureCtx.type()))
        U.dumpThreads(log, !failureTypeIgnored(failureCtx, hnd));
    DiagnosticProcessor diagnosticProcessor = ctx.diagnostic();
    if (diagnosticProcessor != null)
        diagnosticProcessor.onFailure(failureCtx);
    boolean invalidated = hnd.onFailure(ignite, failureCtx);
    if (invalidated) {
        this.failureCtx = failureCtx;
        log.error("Ignite node is in invalid state due to a critical failure.");
    }
    return invalidated;
}

Example 3 with CorruptedDataStructureException

use of org.apache.ignite.internal.processors.cache.persistence.CorruptedDataStructureException in project ignite by apache.

the class BPlusTree method findFirst.

/**
 * Returns a value mapped to the lowest key, or {@code null} if tree is empty or no entry matches the passed filter.
 * @param filter Filter closure.
 * @return Value.
 * @throws IgniteCheckedException If failed.
 */
public T findFirst(TreeRowClosure<L, T> filter) throws IgniteCheckedException {
    checkDestroyed();
    long curPageId = 0L;
    long nextPageId = 0L;
    try {
        for (; ; ) {
            long metaPage = acquirePage(metaPageId);
            try {
                // Level 0 is always at the bottom.
                curPageId = getFirstPageId(metaPageId, metaPage, 0);
            } finally {
                releasePage(metaPageId, metaPage);
            }
            long curPage = acquirePage(curPageId);
            try {
                long curPageAddr = readLock(curPageId, curPage);
                if (curPageAddr == 0)
                    // The first page has gone: restart scan.
                    continue;
                try {
                    BPlusIO<L> io = io(curPageAddr);
                    assert io.isLeaf();
                    for (; ; ) {
                        int cnt = io.getCount(curPageAddr);
                        for (int i = 0; i < cnt; ++i) {
                            if (filter == null || filter.apply(this, io, curPageAddr, i))
                                return getRow(io, curPageAddr, i);
                        }
                        nextPageId = io.getForward(curPageAddr);
                        if (nextPageId == 0)
                            return null;
                        long nextPage = acquirePage(nextPageId);
                        try {
                            long nextPageAddr = readLock(nextPageId, nextPage);
                            // In the current implementation the next page can't change when the current page is locked.
                            assert nextPageAddr != 0 : nextPageAddr;
                            try {
                                long pa = curPageAddr;
                                // Set to zero to avoid double unlocking in finalizer.
                                curPageAddr = 0;
                                readUnlock(curPageId, curPage, pa);
                                long p = curPage;
                                // Set to zero to avoid double release in finalizer.
                                curPage = 0;
                                releasePage(curPageId, p);
                                curPageId = nextPageId;
                                curPage = nextPage;
                                curPageAddr = nextPageAddr;
                                nextPage = 0;
                                nextPageAddr = 0;
                            } finally {
                                if (nextPageAddr != 0)
                                    readUnlock(nextPageId, nextPage, nextPageAddr);
                            }
                        } finally {
                            if (nextPage != 0)
                                releasePage(nextPageId, nextPage);
                        }
                    }
                } finally {
                    if (curPageAddr != 0)
                        readUnlock(curPageId, curPage, curPageAddr);
                }
            } finally {
                if (curPage != 0)
                    releasePage(curPageId, curPage);
            }
        }
    } catch (CorruptedDataStructureException e) {
        throw e;
    } catch (IgniteCheckedException e) {
        throw new IgniteCheckedException("Runtime failure on first row lookup", e);
    } catch (RuntimeException | AssertionError e) {
        throw corruptedTreeException("Runtime failure on first row lookup", e, grpId, curPageId, nextPageId);
    } finally {
        checkDestroyed();
    }
}

Example 4 with CorruptedDataStructureException

use of org.apache.ignite.internal.processors.cache.persistence.CorruptedDataStructureException in project ignite by apache.

the class BPlusTree method findOne.

/**
 * @param row Lookup row for exact match.
 * @param x Implementation specific argument, {@code null} always means that we need to return full detached data row.
 * @return Found result or {@code null}.
 * @throws IgniteCheckedException If failed.
 */
public final <R> R findOne(L row, TreeRowClosure<L, T> c, Object x) throws IgniteCheckedException {
    checkDestroyed();
    GetOne g = new GetOne(row, c, x, false);
    try {
        doFind(g);
        return (R) g.row;
    } catch (CorruptedDataStructureException e) {
        throw e;
    } catch (IgniteCheckedException e) {
        throw new IgniteCheckedException("Runtime failure on lookup row: " + row, e);
    } catch (RuntimeException | AssertionError e) {
        throw corruptedTreeException("Runtime failure on lookup row: " + row, e, grpId, g.pageId);
    } finally {
        checkDestroyed();
    }
}

Example 5 with CorruptedDataStructureException

use of org.apache.ignite.internal.processors.cache.persistence.CorruptedDataStructureException in project ignite by apache.

the class BPlusTree method doPut.

/**
 * @param row New value.
 * @param needOld {@code True} If need return old value.
 * @return Old row.
 * @throws IgniteCheckedException If failed.
 */
private T doPut(T row, boolean needOld) throws IgniteCheckedException {
    checkDestroyed();
    Put p = new Put(row, needOld);
    try {
        for (; ; ) {
            // Go down with retries.
            p.init();
            Result res = putDown(p, p.rootId, 0L, p.rootLvl);
            switch(res) {
                case RETRY:
                case RETRY_ROOT:
                    checkInterrupted();
                    continue;
                case FOUND:
                    // We may need to perform an inner replace on the upper level.
                    if (!p.isFinished()) {
                        res = p.finishTail();
                        // If not found, then the root split has happened and operation should be retried from the actual root.
                        if (res == RETRY || res == NOT_FOUND) {
                            p.releaseTail();
                            assert p.checkTailLevel(getRootLevel()) : "tail=" + p.tail + ", res=" + res;
                            checkInterrupted();
                            continue;
                        }
                    }
                    return p.oldRow;
                default:
                    throw new IllegalStateException("Result: " + res);
            }
        }
    } catch (CorruptedDataStructureException e) {
        throw e;
    } catch (IgniteCheckedException e) {
        throw new IgniteCheckedException("Runtime failure on row: " + row, e);
    } catch (RuntimeException | AssertionError e) {
        throw corruptedTreeException("Runtime failure on row: " + row, e, grpId, p.pageId);
    } finally {
        checkDestroyed();
    }
}