gem-hunter

Current version: 1.5 (amcheck_next extension/SQL version: 2)

Author: Peter Geoghegan <pg@bowt.ie>

License: PostgreSQL license

Note that Microsoft Windows is supported, but only on point releases that have the necessary workaround for various restrictions on dynamic linking that only exist on that platform. The minimum supported point releases are 9.4.16, 9.5.11, 9.6.7, and 10.2.

The amcheck module provides functions that allow you to verify the logical consistency of the structure of PostgreSQL indexes. If the structure appears to be valid, no error is raised. Currently, only B-Tree indexes are supported, although since in practice the majority of PostgreSQL indexes are B-Tree indexes, amcheck is likely to be effective as a general corruption smoke-test in production PostgreSQL installations.

See Using amcheck effectively for information about the kinds of real-world problems amcheck is intended to detect.

amcheck is a contrib extension module that originally appeared in PostgreSQL 10. This externally maintained version of the extension, which is formally named amcheck_next to avoid conflicts with contrib/amcheck, provides the same functionality as PostgreSQL 11's contrib/amcheck to earlier versions of PostgreSQL.

It is safe (though generally not useful) to install amcheck_next alongside contrib/amcheck.

amcheck provides functions that specifically verify various invariants in the structure of the representation of particular indexes. The correctness of the access method functions behind index scans and other important operations relies on these invariants always holding. For example, certain functions verify, among other things, that all B-Tree pages have items in "logical", sorted order (e.g., for B-Tree indexes on text, index tuples should be in collated lexical order). If that particular invariant somehow fails to hold, we can expect binary searches on the affected page to incorrectly guide index scans, resulting in wrong answers to SQL queries.

Verification is performed using the same procedures as those used by index scans themselves, which may be user-defined operator class code. For example, B-Tree index verification relies on comparisons made with one or more B-Tree support function 1 routines, much like B-Tree index scans rely on the routines to guide the scan to a point in the underlying table; see the PostgreSQL documentation on Index Access Methods and Operator Classes for details of operator class support functions.

It is recommended that amcheck be installed using prebuilt packages where available.

The most recent amcheck release is available from the PostgreSQL Community APT repository. Setup instructions can be found in the APT section of the PostgreSQL Wiki.

Once the Community APT repository is set up, and PostgreSQL has itself been installed from a community package, installation of amcheck is generally a simple matter of installing the package that matches your PostgreSQL version:

sudo aptitude install postgresql-10-amcheck

The most recent amcheck release is available from the PostgreSQL Community yum repository. Setup can be performed by following the PostgreSQL yum Howto.

Once the Community yum repository is set up, and PostgreSQL has itself been installed from a community package, installation of amcheck is generally a simple matter of installing the package that matches your PostgreSQL version:

sudo yum install amcheck_next10

The module can be built using the standard PGXS infrastructure. For this to work, you will need to have the pg_config program available in your $PATH.

If you are using a packaged PostgreSQL build and have pg_config available (and in your OS user's $PATH), the procedure is as follows:

tar xvzf amcheck-1.5.tar.gz
cd amcheck-1.5
make
make install

Note that just because pg_config is located in one user's $PATH does not necessarily make it so for the root user.

The Makefile also provides a target for building Debian packages. The target has a dependency on debhelper, devscripts, postgresql-server-dev-all, and the PostgreSQL source package itself (e.g. postgresql-server-dev-9.4).

The packages can be created and installed from the amcheck directory as follows:

sudo aptitude install debhelper devscripts postgresql-server-dev-all
make deb
sudo dpkg -i ./build/postgresql-9.4-amcheck_*.deb

Once the module is built and/or installed, it may be created as a PostgreSQL extension:

mydb=# CREATE EXTENSION amcheck_next;

amcheck functions may be used only by superusers.

The amcheck_next extension has a simple interface. amcheck_next consists of just a few functions that can be used for verification of a named B-Tree index. Note that currently, no function inspects the structure of the underlying heap representation (table).

regclass function arguments are used by amcheck to identify particular index relations. This allows amcheck to accept arguments using various SQL calling conventions:

  -- Use string literal regclass input:
  SELECT bt_index_check('pg_database_oid_index', true);
  -- Use oid regclass input (both perform equivalent verification):
  SELECT bt_index_check(2672, false);
  SELECT bt_index_check(oid, false) FROM pg_class
  WHERE relname = 'pg_database_oid_index';

See the PostgreSQL documentation on Object identifier types for more information.

bt_index_check(index regclass, heapallindexed boolean DEFAULT false)
returns void

bt_index_check tests that its target, a B-Tree index, respects a variety of invariants. Example usage:

  SELECT bt_index_check(index => c.oid, heapallindexed => i.indisunique),
          c.relname,
          c.relpages
  FROM pg_index i
  JOIN pg_opclass op ON i.indclass[0] = op.oid
  JOIN pg_am am ON op.opcmethod = am.oid
  JOIN pg_class c ON i.indexrelid = c.oid
  JOIN pg_namespace n ON c.relnamespace = n.oid
  WHERE am.amname = 'btree' AND n.nspname = 'pg_catalog'
  -- Don't check temp tables, which may be from another session:
  AND c.relpersistence != 't'
  -- Function may throw an error when this is omitted:
  AND c.relkind = 'i' AND i.indisready AND i.indisvalid
  ORDER BY c.relpages DESC LIMIT 10;

   bt_index_check |             relname             | relpages
  ----------------+---------------------------------+----------
                  | pg_depend_reference_index       |       43
                  | pg_depend_depender_index        |       40
                  | pg_proc_proname_args_nsp_index  |       31
                  | pg_description_o_c_o_index      |       21
                  | pg_attribute_relid_attnam_index |       14
                  | pg_proc_oid_index               |       10
                  | pg_attribute_relid_attnum_index |        9
                  | pg_amproc_fam_proc_index        |        5
                  | pg_amop_opr_fam_index           |        5
                  | pg_amop_fam_strat_index         |        5

This example shows a session that performs verification of catalog indexes. Verification of the presence of heap tuples as index tuples is requested for unique indexes only. Since no error is raised, all indexes tested appear to be logically consistent. Naturally, this query could easily be changed to call bt_index_check for every index in the database where verification is supported. An AccessShareLock is acquired on the target index and heap relation by bt_index_check. This lock mode is the same lock mode acquired on relations by simple SELECT statements.

bt_index_check does not verify invariants that span child/parent relationships, but will verify the presence of all heap tuples as index tuples within the index when heapallindexed is true. When a routine, lightweight test for corruption is required in a live production environment, using bt_index_check often provides the best trade-off between thoroughness of verification and limiting the impact on application performance and availability.

bt_index_parent_check(index regclass, heapallindexed boolean DEFAULT false)
returns void

bt_index_parent_check tests that its target, a B-Tree index, respects a variety of invariants. Optionally, when the heapallindexed argument is true, the function verifies the presence of all heap tuples that should be found within the index, and that there are no missing downlinks in the index structure. The checks performed by bt_index_parent_check are a superset of the checks performed by bt_index_check when called with the same options. bt_index_parent_check can be thought of as a more thorough variant of bt_index_check: unlike bt_index_check, bt_index_parent_check also checks invariants that span parent/child relationships. bt_index_parent_check follows the general convention of raising an error if it finds a logical inconsistency or other problem.

A ShareLock is required on the target index by bt_index_parent_check (a ShareLock is also acquired on the heap relation). These locks prevent concurrent data modification from INSERT, UPDATE, and DELETE commands. The locks also prevent the underlying relation from being concurrently processed by VACUUM (and other utility commands). Note that the function holds locks for as short a duration as possible, so there is no advantage to verifying each index individually in a series of transactions, unless long running queries happen to be of particular concern.

bt_index_parent_check's additional verification is more likely to detect various pathological cases. These cases may involve an incorrectly implemented B-Tree operator class used by the index that is checked, or, hypothetically, undiscovered bugs in the underlying B-Tree index access method code. Note that bt_index_parent_check cannot be called when Hot Standby is enabled (i.e., on read-only physical replicas), unlike bt_index_check.

When the heapallindexed argument to verification functions is true, an additional phase of verification is performed against the table associated with the target index relation. This consists of a "dummy" CREATE INDEX operation, which checks for the presence of all would-be new index tuples against a temporary, in-memory summarizing structure (this is built when needed during the first, standard phase). The summarizing structure "fingerprints" every tuple found within the target index. The high level principle behind heapallindexed verification is that a new index that is equivalent to the existing, target index must only have entries that can be found in the existing structure.

The additional heapallindexed phase adds significant overhead: verification will typically take several times longer than it would with only the standard consistency checking of the target index's structure. However, verification will still take significantly less time than an actual CREATE INDEX. There is no change to the relation-level locks acquired when heapallindexed verification is performed. The summarizing structure is bound in size by maintenance_work_mem. In order to ensure that there is no more than a 2% probability of failure to detect the absence of any particular index tuple, approximately 2 bytes of memory are needed per index tuple. As less memory is made available per index tuple, the probability of missing an inconsistency increases. This is considered an acceptable trade-off, since it limits the overhead of verification very significantly, while only slightly reducing the probability of detecting a problem, especially for installations where verification is treated as a routine maintenance task.

With many databases, even the default maintenance_work_mem setting of 64MB is sufficient to have less than a 2% probability of overlooking any single absent or corrupt tuple. This will be the case when there are no indexes with more than about 30 million distinct index tuples, regardless of the overall size of any index, the total number of indexes, or anything else. False positive candidate tuple membership tests within the summarizing structure occur at random, and are very unlikely to be the same for repeat verification operations. Furthermore, within a single verification operation, each missing or malformed index tuple independently has the same chance of being detected. If there is any inconsistency at all, it isn't particularly likely to be limited to a single tuple. All of these factors favor accepting a limited per operation per tuple probability of missing corruption, in order to enable performing more thorough index to heap verification more frequently (practical concerns about the overhead of verification are likely to limit the frequency of verification). In aggregate, the probability of detecting a hardware fault or software defect actually increases significantly with this strategy in most real world cases. Moreover, frequent verification allows problems to be caught earlier on average, which helps to limit the overall impact of corruption, and often simplifies root cause analysis.

amcheck can be effective at detecting various types of failure modes that data page checksums will always fail to catch. These include:

• Structural inconsistencies caused by incorrect operator class implementations.

This includes issues caused by the comparison rules of operating system collations changing. Comparisons of datums of a collatable type like text must be immutable (just as all comparisons used for B-Tree index scans must be immutable), which implies that operating system collation rules must never change.

Though rare, updates to operating system collation rules can cause these issues. More commonly, an inconsistency in the collation order between a master server and a standby server is implicated, possibly because the major operating system version in use is inconsistent. Such inconsistencies will generally only arise on standby servers, and so can generally only be detected on standby servers.

If a problem like this arises, it may not affect each individual index that is ordered using an affected collation, simply because indexed values might happen to have the same absolute ordering regardless of the behavioral inconsistency.

• Structural inconsistencies between indexes and the heap relations that are indexed (when heapallindexed verification is performed).

There is no cross-checking of indexes against their heap relation during normal operation. Symptoms of heap corruption can be very subtle.

• Corruption caused by hypothetical undiscovered bugs in the underlying PostgreSQL access method code, sort code, or transaction management code.

Automatic verification of the structural integrity of indexes plays a role in the general testing of new or proposed PostgreSQL features that could plausibly allow a logical inconsistency to be introduced. Verification of table structure and associated visibility and transaction status information plays a similar role. One obvious testing strategy is to call amcheck functions continuously when running the standard regression tests.

• Filesystem or storage subsystem faults where checksums happen to simply not be enabled.

Note that amcheck examines a page as represented in some shared memory buffer at the time of verification if there is only a shared buffer hit when accessing the block. Consequently, amcheck does not necessarily examine data read from the filesystem at the time of verification. Note that when checksums are enabled, amcheck may raise an error due to a checksum failure when a corrupt block is read into a buffer.

• Corruption caused by faulty RAM, and the broader memory subsystem and operating system.

PostgreSQL does not protect against correctable memory errors and it is assumed you will operate using RAM that uses industry standard Error Correcting Codes (ECC) or better protection. However, ECC memory is typically only immune to single-bit errors, and should not be assumed to provide absolute protection against failures that result in memory corruption.

When heapallindexed verification is performed, there is generally a greatly increased chance of detecting single-bit errors, since strict binary equality is tested, and the indexed attributes within the heap are tested.

The overhead of calling bt_index_check for every index on a live production system is roughly comparable to the overhead of vacuuming; like VACUUM, verification uses a "buffer access strategy", which limits its impact on which pages are cached within shared_buffers. A major design goal of amcheck is to support routine verification of all indexes on busy production systems. Note that heapallindexed verification significantly increases the runtime of verification.

No amcheck routine will ever modify data, and so no pages will ever be "dirtied", which is not the case with VACUUM. On the other hand, amcheck may be required to verify a large number of indexes all at once, which is typically not a behavior that autovacuum exhibits. amcheck exhaustively accesses every page in each index verified. This behavior is useful in part because verification may detect a checksum failure, which may have previously gone undetected only because no process needed data from the corrupt page in question, including even an autovacuum worker process.

Note also that bt_index_check and bt_index_parent_check access the contents of indexes in "logical" order, which, in the worst case, implies that all I/O operations are performed at random positions on the filesystem. In contrast, VACUUM always removes dead index tuples from B-Tree indexes while accessing the contents of B-Tree indexes in sequential order.

No error concerning corruption raised by amcheck should ever be a false positive. amcheck raises errors in the event of conditions that, by definition, should never happen. It seems unlikely that there could ever be a useful general remediation to problems it detects.

In general, an explanation for the root cause of an invariant violation should be sought. contrib/pageinspect can play a useful role in diagnosing corruption that amcheck highlights. A REINDEX may or may not be effective in repairing corruption, depending on the exact details of how the corruption originated.

In general, amcheck can only prove the presence of corruption; it cannot prove its absence.