- Measure size of redo WAL (new histogram), with bounds between 24B-32kB
- Add 2 more buckets at the upper end of the redo time histogram
We often (>0.1% of several hours each day) take more than 250ms to do the
redo round-trip to the postgres process. We need to measure these redo
times more precisely.
- Split postgres_ffi into two version specific files.
- Preserve pg_version in timeline metadata.
- Use pg_version in safekeeper code. Check for postgres major version mismatch.
- Clean up the code to use DEFAULT_PG_VERSION constant everywhere, instead of hardcoding.
- Parameterize python tests: use DEFAULT_PG_VERSION env and pg_version fixture.
To run tests using a specific PostgreSQL version, pass the DEFAULT_PG_VERSION environment variable:
'DEFAULT_PG_VERSION='15' ./scripts/pytest test_runner/regress'
Currently don't all tests pass, because rust code relies on the default version of PostgreSQL in a few places.
* Add test for pageserver metric cleanup once a tenant is detached.
* Remove tenant specific timeline metrics on detach.
* Use definitions from timeline_metrics in page service.
* Move metrics to own file from layered_repository/timeline.rs
* TIMELINE_METRICS: define smgr metrics
* REMOVE SMGR cleanup from timeline_metrics. Doesn't seem to work as
expected.
* Vritual file centralized metrics, except for evicted file as there's no
tenat id or timeline id.
* Use STORAGE_TIME from timeline_metrics in layered_repository.
* Remove timelineless gc metrics for tenant on detach.
* Rename timeline metrics -> metrics as it's more generic.
* Don't create a TimelineMetrics instance for VirtualFile
* Move the rest of the metric definitions to metrics.rs too.
* UUID -> ZTenantId
* Use consistent style for dict.
* Use Repository's Drop trait for dropping STORAGE_TIME metrics.
* No need for Arc, TimelineMetrics is used in just one place. Due to that,
we can fall back using ZTenantId and ZTimelineId too to avoid additional
string allocation.
Previously, it could only distinguish REDO task durations down to 5ms, which
equates to approx. 200pages/sec or 1.6MB/sec getpage@LSN traffic.
This patch improves to 200'000 pages/sec or 1.6GB/sec, allowing for
much more precise performance measurement of the redo process.
Re-export only things that are used by other modules.
In the future, I'm imagining that we run bindgen twice, for Postgres
v14 and v15. The two sets of bindings would go into separate
'bindings_v14' and 'bindings_v15' modules.
Rearrange postgres_ffi modules.
Move function, to avoid Postgres version dependency in timelines.rs
Move function to generate a logical-message WAL record to postgres_ffi.
* Do not create initial tenant and timeline (adjust Python tests for that)
* Rework config handling during init, add --update-config to manage local config updates
Try to follow Prometheus style-guide https://prometheus.io/docs/practices/naming/ for metrics names. More specifically:
- Use `pageserver_` prefix for all pagserver metrics
- Specify `_seconds` unit in time metrics
- Use unit as a suffix in other cases, such as `_hits`, `_bytes`, `_records`
- Use `_total` suffix for accumulating counters (note that Histograms append that suffix internally)
A new `get_lsn_by_timestamp` command is added to the libpq page service
API.
An extra timestamp field is now stored in an extra field after each
Clog page. It is the timestamp of the latest commit, among all the
transactions on the Clog page. To find the overall latest commit, we
need to scan all Clog pages, but this isn't a very frequent operation
so that's not too bad.
To find the LSN that corresponds to a timestamp, we perform a binary
search. The binary search starts with min = last LSN when GC ran, and
max = latest LSN on the timeline. On each iteration of the search we
check if there are any commits with a higher-than-requested timestamp
at that LSN.
Implements github issue 1361.
This is a backwards-incompatible change. The new pageserver cannot
read repositories created with an old pageserver binary, or vice
versa.
Simplify Repository to a value-store
------------------------------------
Move the responsibility of tracking relation metadata, like which
relations exist and what are their sizes, from Repository to a new
module, pgdatadir_mapping.rs. The interface to Repository is now a
simple key-value PUT/GET operations.
It's still not any old key-value store though. A Repository is still
responsible from handling branching, and every GET operation comes
with an LSN.
Mapping from Postgres data directory to keys/values
---------------------------------------------------
All the data is now stored in the key-value store. The
'pgdatadir_mapping.rs' module handles mapping from PostgreSQL objects
like relation pages and SLRUs, to key-value pairs.
The key to the Repository key-value store is a Key struct, which
consists of a few integer fields. It's wide enough to store a full
RelFileNode, fork and block number, and to distinguish those from
metadata keys.
'pgdatadir_mapping.rs' is also responsible for maintaining a
"partitioning" of the keyspace. Partitioning means splitting the
keyspace so that each partition holds a roughly equal number of keys.
The partitioning is used when new image layer files are created, so
that each image layer file is roughly the same size.
The partitioning is also responsible for reclaiming space used by
deleted keys. The Repository implementation doesn't have any explicit
support for deleting keys. Instead, the deleted keys are simply
omitted from the partitioning, and when a new image layer is created,
the omitted keys are not copied over to the new image layer. We might
want to implement tombstone keys in the future, to reclaim space
faster, but this will work for now.
Changes to low-level layer file code
------------------------------------
The concept of a "segment" is gone. Each layer file can now store an
arbitrary range of Keys.
Checkpointing, compaction
-------------------------
The background tasks are somewhat different now. Whenever
checkpoint_distance is reached, the WAL receiver thread "freezes" the
current in-memory layer, and creates a new one. This is a quick
operation and doesn't perform any I/O yet. It then launches a
background "layer flushing thread" to write the frozen layer to disk,
as a new L0 delta layer. This mechanism takes care of durability. It
replaces the checkpointing thread.
Compaction is a new background operation that takes a bunch of L0
delta layers, and reshuffles the data in them. It runs in a separate
compaction thread.
Deployment
----------
This also contains changes to the ansible scripts that enable having
multiple different pageservers running at the same time in the staging
environment. We will use that to keep an old version of the pageserver
running, for clusters created with the old version, at the same time
with a new pageserver with the new binary.
Author: Heikki Linnakangas
Author: Konstantin Knizhnik <knizhnik@zenith.tech>
Author: Andrey Taranik <andrey@zenith.tech>
Reviewed-by: Matthias Van De Meent <matthias@zenith.tech>
Reviewed-by: Bojan Serafimov <bojan@zenith.tech>
Reviewed-by: Konstantin Knizhnik <knizhnik@zenith.tech>
Reviewed-by: Anton Shyrabokau <antons@zenith.tech>
Reviewed-by: Dhammika Pathirana <dham@zenith.tech>
Reviewed-by: Kirill Bulatov <kirill@zenith.tech>
Reviewed-by: Anastasia Lubennikova <anastasia@zenith.tech>
Reviewed-by: Alexey Kondratov <alexey@zenith.tech>
If a heap UPDATE record modified two pages, and both pages needed to have
their VM bits cleared, and the VM bits were located on the same VM page,
we would emit two ZenithWalRecord::ClearVisibilityMapFlags records for
the same VM page. That produced warnings like this in the pageserver log:
Page version Wal(ClearVisibilityMapFlags { heap_blkno: 18, flags: 3 }) of rel 1663/13949/2619_vm blk 0 at 2A/346046A0 already exists
To fix, change ClearVisibilityMapFlags so that it can update the bits
for both pages as one operation.
This was already covered by several python tests, so no need to add a
new one. Fixes#1125.
Co-authored-by: Konstantin Knizhnik <knizhnik@zenith.tech>
Introduce the concept of a "ZenithWalRecord", which can be a Postgres WAL
record that is replayed with the Postgres WAL redo process, or a built-in
type that is handled entirely by pageserver code.
Replace the special code to replay Postgres XACT commit/abort records
with new Zenith WAL records. A separate zenith WAL record is created for
each modified CLOG page. This allows removing the 'main_data_offset'
field from stored PostgreSQL WAL records, which saves some memory and
some disk space in delta layers.
Introduce zenith WAL records for updating bits in the visibility map.
Previously, when e.g. a heap insert cleared the VM bit, we duplicated the
heap insert WAL record for the affected VM page. That was very wasteful.
The heap WAL record could be massive, containing a full page image in
the worst case. This addresses github issue #941.
If a commit record contains XIDs that are stored on different CLOG pages,
we duplicate the commit record for each affected CLOG page. In the redo
routine, we must only apply the parts of the record that apply to the
CLOG page being restored. We got that right in the loop that handles the
sub-XIDs, but incorrectly always set the bit that corresponds to the main
XID.
The logic to compute the page number was broken, and as a result, only
the first page of multixact members was updated correctly. All the
rest were left as zeros. Improve test_multixact.py to generate more
multixacts, to cover this case.
Also fix the check that the restored PG data directory matches the
original one. Previously, the test compared the 'pg_new' cluster,
which is a bit silly because the test restored the 'pg_new' cluster
only a few lines earlier, so if the multixact WAL redo is somehow
broken, the comparison will just compare two broken data directories
and report success. Change it to compare the original datadir, the one
where the multixacts were originally created, with a restored image of
the same.
Move the code for decoding a WAL stream into WAL records into
'postgres_ffi', and keep the code to parse the WAL records deeper in
'pageserver' crate, renamed to walrecord.rs.
This tidies up the dependencies a bit. 'walkeeper' reuses the same
waldecoder routines, and it used to depend on 'pageserver' because of
that. Now it only depends on 'postgres_ffi'.
(The comment in walkeeper/Cargo.toml that claimed that the dependency was
needed for ZTimelineId was obsolete. ZTimelineId is defined in
'zenith_utils', the dependency was actually needed for the waldecoder.)
The tokio futures added some overhead, so switch to plain non-blocking
I/O with poll(). In a simple pgbench test on my laptop (select-only
queries, scale-factor 1 `pgbench -P1 -T50 -S`), this gives about 10%
improvement, from about 4300 TPS to 4800 TPS.
Instead of building a separate Vec<u8> to hold each message, serialize all
the messages to one big Vec<u8>. This eliminates some Vec allocation and
memcpy() overhead. The downside is that if there are a lot of records to
replay, we have to serialize them all into one big chunk of memory.
That shouldn't be a problem in practice. If you need to replay millions
of records to reconstruct a page, we should've materialized a new image
of that page earlier already.
No need to use BytesMut in these functions. Plain Vec is simpler. And
should be marginally faster too; I saw BytesMut functions previously
in 'perf' profile, consuming around 5% of the overall pageserver CPU
time. That's gone with this patch, although I don't see any discernible
difference in the overall performance test results.
Reduces the CPU time spent in the write() syscalls. I noticed that we were
spending a lot of CPU time in libc::write, coming from request_redo(), in
the 'bulk_insert' test. According to some quick profiling with 'perf',
this reduces the CPU time spent in request_redo() from about 30% to 15%.
For some reason, it doesn't reduce the overall runtime of the 'bulk_insert'
test much, maybe by one second if you squint (from about 37s to 36s), so
there must be some other bottleneck, like I/O. But this is surely still
a good idea, just based on the reduced CPU cycles.
Commit message copied below:
* Allow LeSer/BeSer impls missing Serialize/Deserialize
Currently, using `LeSer` or `BeSer` requires that the type implements
both `Serialize` and `DeserializeOwned`, even if we're only using the
trait for one of those functionalities.
Moving the bounds to the methods gives the convenience of the traits
without requiring unnecessary derives.
* Remove unused #[derive(Serialize/Deserialize)]
This should hopefully reduce compile times - if only by a little bit.
Some of these were already unused (we weren't using LeSer/BeSer for the
types), but most are have *become* unused with the change to
LeSer/BeSer.
* Allow LeSer/BeSer impls missing Serialize/Deserialize
Currently, using `LeSer` or `BeSer` requires that the type implements
both `Serialize` and `DeserializeOwned`, even if we're only using the
trait for one of those functionalities.
Moving the bounds to the methods gives the convenience of the traits
without requiring unnecessary derives.
* Remove unused #[derive(Serialize/Deserialize)]
This should hopefully reduce compile times - if only by a little bit.
Some of these were already unused (we weren't using LeSer/BeSer for the
types), but most are have *become* unused with the change to
LeSer/BeSer.
Track the time spent on replaying WAL records by the special Postgres
process, the time spent waiting for acces to the Postgres process (since
there is only one per tenant), and the number of records replayed.
My main motivation is to make it easier to attribute time spent in WAL
redo to the request that needed the WAL redo. With this patch, the WAL
redo is performed by the requester thread, so it shows up in stack traces
and in 'perf' report as part of the requester's call stack. This is also
slightly simpler (less lines of code) and should be a bit faster too.
Follow PostgreSQL logic: remove Twophase files when prepared transaction is committed/aborted.
Always store Twophase segments as materialized page images (no wal records).
Current state with authentication.
Page server validates JWT token passed as a password during connection
phase and later when performing an action such as create branch tenant
parameter of an operation is validated to match one submitted in token.
To allow access from console there is dedicated scope: PageServerApi,
this scope allows access to all tenants. See code for access validation in:
PageServerHandler::check_permission.
Because we are in progress of refactoring of communication layer
involving wal proposer protocol, and safekeeper<->pageserver. Safekeeper
now doesn’t check token passed from compute, and uses “hardcoded” token
passed via environment variable to communicate with pageserver.
Compute postgres now takes token from environment variable and passes it
as a password field in pageserver connection. It is not passed through
settings because then user will be able to retrieve it using pg_settings
or SHOW ..
I’ve added basic test in test_auth.py. Probably after we add
authentication to remaining network paths we should enable it by default
and switch all existing tests to use it.
This clarifies - I hope - the abstractions between Repository and
ObjectRepository. The ObjectTag struct was a mix of objects that could
be accessed directly through the public Timeline interface, and also
objects that were created and used internally by the ObjectRepository
implementation and not supposed to be accessed directly by the
callers. With the RelishTag separaate from ObjectTag, the distinction
is more clear: RelishTag is used in the public interface, and
ObjectTag is used internally between object_repository.rs and
object_store.rs, and it contains the internal metadata object types.
One awkward thing with the ObjectTag struct was that the Repository
implementation had to distinguish between ObjectTags for relations,
and track the size of the relation, while others were used to store
"blobs". With the RelishTags, some relishes are considered
"non-blocky", and the Repository implementation is expected to track
their sizes, while others are stored as blobs. I'm not 100% happy with
how RelishTag captures that either: it just knows that some relish
kinds are blocky and some non-blocky, and there's an is_block()
function to check that. But this does enable size-tracking for SLRUs,
allowing us to treat them more like relations.
This changes the way SLRUs are stored in the repository. Each SLRU
segment, e.g. "pg_clog/0000", "pg_clog/0001", are now handled as a
separate relish. This removes the need for the SLRU-specific
put_slru_truncate() function in the Timeline trait. SLRU truncation is
now handled by caling put_unlink() on the segment. This is more in
line with how PostgreSQL stores SLRUs and handles their trunction.
The SLRUs are "blocky", so they are accessed one 8k page at a time,
and repository tracks their size. I considered an alternative design
where we would treat each SLRU segment as non-blocky, and just store
the whole file as one blob. Each SLRU segment is up to 256 kB in size,
which isn't that large, so that might've worked fine, too. One reason
I didn't do that is that it seems better to have the WAL redo
routines be as close as possible to the PostgreSQL routines. It
doesn't matter much in the repository, though; we have to track the
size for relations anyway, so there's not much difference in whether
we also do it for SLRUs.
While working on this, I noticed that the CLOG and MultiXact redo code
did not handle wraparound correctly. We need to fix that, but for now,
I just commented them out with a FIXME comment.
