/* trail.h : internal interface to backing out of aborted Berkeley DB txns * * ==================================================================== * Licensed to the Apache Software Foundation (ASF) under one * or more contributor license agreements. See the NOTICE file * distributed with this work for additional information * regarding copyright ownership. The ASF licenses this file * to you under the Apache License, Version 2.0 (the * "License"); you may not use this file except in compliance * with the License. You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, * software distributed under the License is distributed on an * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY * KIND, either express or implied. See the License for the * specific language governing permissions and limitations * under the License. * ==================================================================== */ #ifndef SVN_LIBSVN_FS_TRAIL_H #define SVN_LIBSVN_FS_TRAIL_H #define SVN_WANT_BDB #include "svn_private_config.h" #include #include "svn_fs.h" #include "fs.h" #ifdef __cplusplus extern "C" { #endif /* __cplusplus */ /* "How do I get a trail object? All these functions in the filesystem expect them, and I can't find a function that returns one." Well, there isn't a function that returns a trail. All trails come from svn_fs_base__retry_txn. Here's how to use that: When using Berkeley DB transactions to protect the integrity of a database, there are several things you need to keep in mind: - Any Berkeley DB operation you perform as part of a Berkeley DB transaction may return DB_LOCK_DEADLOCK, meaning that your operation interferes with some other transaction in progress. When this happens, you must abort the transaction, which undoes all the changes you've made so far, and try it again. So every piece of code you ever write to bang on the DB needs to be wrapped up in a retry loop. - If, while you're doing your database operations, you also change some in-memory data structures, then you may want to revert those changes if the transaction deadlocks and needs to be retried. - If you get a `real' error (i.e., something other than DB_LOCK_DEADLOCK), you must abort your DB transaction, to release its locks and return the database to its previous state. Similarly, you may want to unroll some changes you've made to in-memory data structures. - Since a transaction insulates you from database changes made by other processes, it's often possible to cache information about database contents while the transaction lasts. However, this cache may become stale once your transaction is over. So you may need to clear your cache once the transaction completes, either successfully or unsuccessfully. The `svn_fs_base__retry_txn' function and its friends help you manage some of that, in one nice package. To use it, write your code in a function like this: static svn_error_t * txn_body_do_my_thing (void *baton, trail_t *trail) { ... Do everything which needs to be protected by a Berkeley DB transaction here. Use TRAIL->db_txn as your Berkeley DB transaction, and do your allocation in TRAIL->pool. Pass TRAIL on through to any functions which require one. If a Berkeley DB operation returns DB_LOCK_DEADLOCK, just return that using the normal Subversion error mechanism (using DB_ERR, for example); don't write a retry loop. If you encounter some other kind of error, return it in the normal fashion. ... } Now, call svn_fs_base__retry_txn, passing a pointer to your function as an argument: err = svn_fs_base__retry_txn (fs, txn_body_do_my_thing, baton, pool); This will simply invoke your function `txn_body_do_my_thing', passing BATON through unchanged, and providing a fresh TRAIL object, containing a pointer to the filesystem object, a Berkeley DB transaction and an APR pool -- a subpool of POOL -- you should use. If your function returns a Subversion error wrapping a Berkeley DB DB_LOCK_DEADLOCK error, `svn_fs_base__retry_txn' will abort the trail's Berkeley DB transaction for you (thus undoing any database changes you've made), free the trail's subpool (thus undoing any allocation you may have done), and try the whole thing again with a new trail, containing a new Berkeley DB transaction and pool. If your function returns any other kind of Subversion error, `svn_fs_base__retry_txn' will abort the trail's Berkeley DB transaction, free the subpool, and return your error to its caller. If, heavens forbid, your function actually succeeds, returning SVN_NO_ERROR, `svn_fs_base__retry_txn' commits the trail's Berkeley DB transaction, thus making your DB changes permanent, leaves the trail's pool alone so all the objects it contains are still around (unless you request otherwise), and returns SVN_NO_ERROR. Keep the amount of work done in a trail small. C-Mike Pilato said to me: I want to draw your attention to something that you may or may not realize about designing for the BDB backend. The 'trail' objects are (generally) representative of Berkeley DB transactions -- that part I'm sure you know. But you might not realize the value of keeping transactions as small as possible. Berkeley DB will accumulate locks (which I believe are page-level, not as tight as row-level like you might hope) over the course of a transaction, releasing those locks only at transaction commit/abort. Berkeley DB backends are configured to have a maximum number of locks and lockers allowed, and it's easier than you might think to hit the max-locks thresholds (especially under high concurrency) and see an error (typically a "Cannot allocate memory") result from that. For example, in [a loop] you are writing a bunch of rows to the `changes' table. Could be 10. Could be 100,000. 100,000 writes and associated locks might be a problem or it might not. But I use it as a way to encourage you to think about reducing the amount of work you spend in any one trail [...]. */ struct trail_t { /* A Berkeley DB transaction. */ DB_TXN *db_txn; /* The filesystem object with which this trail is associated. */ svn_fs_t *fs; /* A pool to allocate things in as part of that transaction --- a subpool of the one passed to `begin_trail'. We destroy this pool if we abort the transaction, and leave it around otherwise. */ apr_pool_t *pool; #if defined(SVN_FS__TRAIL_DEBUG) struct trail_debug_t *trail_debug; #endif }; typedef struct trail_t trail_t; /* Try a Berkeley DB transaction repeatedly until it doesn't deadlock. That is: - Begin a new Berkeley DB transaction, DB_TXN, in the filesystem FS. - Allocate a subpool of POOL, TXN_POOL. - Start a new trail, TRAIL, pointing to DB_TXN and TXN_POOL. - Apply TXN_BODY to BATON and TRAIL. TXN_BODY should try to do some series of DB operations which needs to be atomic, using TRAIL->db_txn as the transaction, and TRAIL->pool for allocation. If a DB operation deadlocks, or if any other kind of error happens, TXN_BODY should simply return with an appropriate svn_error_t, E. - If TXN_BODY returns SVN_NO_ERROR, then commit the transaction, run any completion functions, and return SVN_NO_ERROR. Do *not* free TXN_POOL (unless DESTROY_TRAIL_POOL is set). - If E is a Berkeley DB error indicating that a deadlock occurred, abort the DB transaction and free TXN_POOL. Then retry the whole thing from the top. - If E is any other kind of error, free TXN_POOL and return E. One benefit of using this function is that it makes it easy to ensure that whatever transactions a filesystem function starts, it either aborts or commits before it returns. If we don't somehow complete all our transactions, later operations could deadlock. */ svn_error_t * svn_fs_base__retry_txn(svn_fs_t *fs, svn_error_t *(*txn_body)(void *baton, trail_t *trail), void *baton, svn_boolean_t destroy_trail_pool, apr_pool_t *pool); svn_error_t * svn_fs_base__retry_debug(svn_fs_t *fs, svn_error_t *(*txn_body)(void *baton, trail_t *trail), void *baton, svn_boolean_t destroy_trail_pool, apr_pool_t *pool, const char *txn_body_fn_name, const char *filename, int line); #if defined(SVN_FS__TRAIL_DEBUG) #define svn_fs_base__retry_txn(fs, txn_body, baton, destroy, pool) \ svn_fs_base__retry_debug(fs, txn_body, baton, destroy, pool, \ #txn_body, __FILE__, __LINE__) #endif /* Try an action repeatedly until it doesn't deadlock. This is exactly like svn_fs_base__retry_txn() (whose documentation you really should read) except that no Berkeley DB transaction is created. */ svn_error_t *svn_fs_base__retry(svn_fs_t *fs, svn_error_t *(*txn_body)(void *baton, trail_t *trail), void *baton, svn_boolean_t destroy_trail_pool, apr_pool_t *pool); /* Record that OPeration is being done on TABLE in the TRAIL. */ #if defined(SVN_FS__TRAIL_DEBUG) void svn_fs_base__trail_debug(trail_t *trail, const char *table, const char *op); #else #define svn_fs_base__trail_debug(trail, table, operation) #endif #ifdef __cplusplus } #endif /* __cplusplus */ #endif /* SVN_LIBSVN_FS_TRAIL_H */