Now that we can have several threads doing COPY, each of them need to
know about the *pgsql-reserved-keywords* list. Make sure that's the case
and in passing fix some call sites to apply-identifier-case.
Also, more disturbingly, fix the code so that TRUNCATE is called from
the main thread before giving control to the COPY threads, rather than
having two concurrent threads doing the TRUNCATE twice. It's rather
strange that we got no complaint from the field on that part...
When the option "WITH no foreign keys" is in use, it's not necessary to
go read the foreign key information_schema bits at all, so just don't
issue the query, and same thing with the "create no indexes" option.
In no that old versions of MySQL, the referential_constraints table of
information_schema doesn't exist, so this should make pgloader
compatible with MySQL 5.0 something and earlier.
It turns out the summary write times included time spent waiting for
batches to be ready, which isn't fair to PostgreSQL COPY implementation,
and moreover doesn't help figuring out the bottlenecks...
Make batches of raw data straight from the reader output (map-rows) and
have the transformation worker focus on changing the batch content from
raw rows to copy strings.
Also review the organisation of responsabilities in the code, allowing
to move queue.lisp into utils/batch.lisp, renaming it as its scope has
been reduced to only care about preparing batches.
This came out of trying to have multiple workers concurrently processing
the batches from the reader and feeding the hardcoded 2 COPY workers,
but it failed for multiple reasons. All is left as of now is this
cleanup, which seems to be on the faster side of things, which is always
good.
We used to wait for the wrong number of workers, meaning the rest of the
code began running before the indexes where all available. A user report
where one of the indexes takes a very long time to compute made it
obvious.
In passing, also improve reporting of those rendez-vous sections.
It came pretty obvious that the error counting was broken, it happens
that I forgot to pass the information down to the state handling parts
of the code.
In passing improve and fix CSV parse errors counting and fatal errors
reporting.
In some cases, like when client_min_messages is set to debug5,
PostgreSQL might send notification messages to the connecting client
even while opening a connection. Those are still considered WARNINGs by
the Postmodern driver...
Handle those warnings by just printing them out in the pgloader logs,
rather than considering those conditions as hard failures (signaling a
db-connection-error).
The test/errors.load set the search_path to include the 'err' schema,
which is to be created by the test itself. PostgreSQL 9.1 raises an
error where 9.4 and following just accept the setting, and Travis runs a
9.1 PostgreSQL.
Let's just create the schema before-hand so that we can still run tests
against SET search_path from the load file.
Filter the list of tables we migrate directly from the SQLite query,
avoiding to return useless data. To do that, use the LIKE pattern
matching supported by SQLite, where the REGEX operator is only available
when extra features are loaded apparently.
See #310 where filtering out the view still caused errors in the
loading.
It's been proven by Andres Freund benchmarks that the best number of
parallel COPY threads concurrently active against a single table is 2 as
of PostgreSQL current development version (up to 9.5 stable, might still
apply to 9.6 depending on how we might solve the problem).
Henceforth hardcode 2 COPY threads in pgloader. This also has the
advantage that in the presence of lots of bad rows, we should sustain a
better throughtput and not stall completely.
Finally, also improve the MySQL setup to use 8 threads by default, that
is be able to load two tables concurrently, each with 2 COPY workers, a
reader and a transformer thread.
It's all still experimental as far as performances go, next patches
should bring the capability to configure the parallelism wanted from the
command line and load command tho.
Also, other source types will want to benefit from the same
capabilities, it just happens that it's easier to play with MySQL first
for some reasons here.
Next parallelism improvements will allow pgloader to use more than one
COPY thread to load data, with the impact of changing the order of rows
in the database.
Rather than doing a copy out and `diff` of the data just loaded, load
the reference data and do the diff in SQL:
select * from loaded.data
except
select * from expected.data
If such a query returns any row, we know we didn't load what was
expected and the regression test is failing.
This regression testing facility should also allow us to finally add
support for multiple-table regression tests (sqlite, mysql, etc).
Our PostgreSQL driver uses CFFI to load the SSL support from open ssl
and as a result the certificate and key file names should be strings
rather than pathnames. Should fix#308 again...
This fixes#308 by automatically using the PostgreSQL Client Side SSL
files as documented in the following reference:
http://www.postgresql.org/docs/current/static/libpq-ssl.html#LIBPQ-SSL-FILE-USAGE
This uses the Postmodern special support for it. Unfortunately couldn't
test it locally other than it doesn't break non-ssl connections. Pushing
to have user feedback.
That allows the copy-column-list specific method for MySQL to be a
method of the common pgloader.sources::copy-column-list generic
function, and then to be called again when needed.
This fixes an oversight in #41e9eeb and fixes#132 again.
Trying to provide a new one fails with an error, that I missed because I
must have forgotten to `make clean` when adding the previous #+ccl
variant here...
This alone doesn't allow to fix building for CCL but already improves
the situation as reported at #303. Next failure is something I fail to
understand tonight:
Fatal SIMPLE-ERROR:
Compilation failed: In MAKE-DOUBLE-FLOAT: Type declarations violated in (THE FIXNUM 4294967295) in /Users/dim/dev/pgloader/build/quicklisp/local-projects/qmynd/src/common/utilities.lisp
When using Clozure Common Lisp apparently a :absolute directory
component for make-pathname is supposed to contain a single path
component, fix by using parse-native-namestring instead.
In case it's needed, the following spelling seems portable enough:
CL-USER> (uiop:merge-pathnames*
(uiop:make-pathname* :directory '(:relative "pgloader"))
(uiop:make-pathname* :directory '(:absolute "tmp")))
#P"/tmp/pgloader/"
We used to have a reader and a writer cooperating concurrently into
loading the data from the source to PostgreSQL. The tranformation of the
data was then the responsibility of the reader thread.
Measurements showed that the PostgreSQL processes were mostly idle,
waiting for the reader to produce data fast enough.
In this patch we introduce a third worker thread that is responsible for
processing the raw data into pre-formatted batches, allowing the reader
to focus on extracting the data only. We now have two lparallel queues
involved in the processing, the raw queue contains the vectors of raw
data directly, and the processed-queue contains batches of properly
encoded strings for the COPY text protocol.
On the test laptop the performance gain isn't noticeable yet, it might
be that we need much larger data sets to see a gain here. At least the
setup isn't detrimental to performances on smaller data sets.
Next improvements are going to allow more features: specialized batch
retry thread and parallel table copy scheduling for database sources.
Let's also continue caring about performances and play with having
several worker and writer threads for each reader. In later patches.
And some day, too, we will need to make the number of workers a user
defined variable rather than something hard coded as today. It's on the
todo list, meanwhile, dear user, consider changing the (make-kernel 6)
into (make-kernel 12) or something else in src/sources/mysql/mysql.lisp,
and consider enlighting me with whatever it is you find by doing so!
Have each thread publish its own start-time so that the main thread may
compute time spent in source and target processing, in order to fix the
crude hack of taking (max read-time write-time) in the total time column
of the summary.
We still have some strange artefacts here: we consider that the full
processing time is bound to the writer thread (:target), because it
needs to have the reader done already to be able to COPY the last
batch... but in testing I've seen some :source timings higher than the
:target ones...
Let's solve problems one at a time tho, I guess multi-threading and
accurate wall clock times aren't to be expected to mix and match that
easily anyway (multi cores, single RTC and all that).
As seen in #302 it's possible to define a SQLite column of type "integer
auto_increment". In my testing tho, it doesn't mean a thing. Worse than
that, apparently when an integer column is created that is also used as
the primary key of the table, the notation "integer auto_increment
primary key" disables the rowid behavior that is certainly expected.
Let's not yet mark the bug as fixed as I suppose we will have to do
something about this rowid mess. Thanks again SQLite.
When calling lparallel:receive-results from the main threads we loose
the ability to measure proper per-table processing times, because it all
happens in parallel and the main threads seems to receive events in the
same millisecond as when the worker is started, meaning it's all 0.0s.
So when we don't have "secs" stats, pick the greatest of read or write
time, which we do have from the worker threads themselves.
The number are still wrong, but less so than the "0.0s" displayed before
this patch.
Now that we can setup many concurrent threads working against the
PostgreSQL database, and before we open the number of workers to our
users, install an heuristic to manage the PostgreSQL error classes “too
many connection” and “configuration limit exceeded” so that pgloader
waits for some time (*retry-connect-delay*) then tries connecting again.
It's quite simplistic but should cover lots of border-line cases way
more nicely than just throwing the interactive debugger at the end user.
The newly added statistics are showing that read+write times are not
enough to explain how long we wait for the data copying, so it must be
the workers setup rather than the workers themselves.
From there, let lparallel work its magic in scheduling the work we do in
parallel in pgloader: rather than doing blocking receive-result calls
for each table, only receive-result at the end of the whole
copy-database processing.
On test data here on the laptop we go from 6s to 3s to migrate the
sakila database from MySQL to PostgreSQL: that's because we have lots of
very small tables, so the cost of waiting after each COPY added up quite
quickly.
In passing, stop sharing the same connection object in between parallel
workers that used to be controlled active in-sequence, see the new API
clone-connection (which takes over new-pgsql-connection).
Add metrics to devise where the time is spent in current pgloader code
so that it's possible to then optimize away the batch processing as we
do it today.
Given the following extract of the measures, it seems that doing the
data transformations in the reader thread isn't so bright an idea. More
to come.
table name total time read write
----------------- -------------- --------- ---------
extract 2.014s
before load 0.050s
fetch 0.000s
----------------- -------------- --------- ---------
geolite.location 16.090s 15.933s 5.732s
geolite.blocks 28.896s 28.795s 5.312s
----------------- -------------- --------- ---------
after load 37.772s
----------------- -------------- --------- ---------
Total import time 1m25.082s 44.728s 11.044s
When devising the common API, the first step has been to implement
specific methods for each generic function of the protocol. It now
appears that in some cases we don't need the extra level of flexibility:
each change of the API has been systematically reported to all the
specific methods, so just use a single generic definition where possible.
In particular, introduce new intermediate class for COPY subclasses
allowing to share more common code in the methods implementation, rather
than having to copy/paste and maintain several versions of the same
code.
It would be good to be able to centralize more code for the database
sources and how they are organized around metadata/import-data/complete
schema, but it doesn't look obvious how to do it just now.
Update the stats used to be a quite simple incf and doing it once per
read row was good enough, but now that it involves sending a message to
the monitor thread let's only send a message per batch, reducing the
communication load here.
The local-time:encode-timestamp function takes a default timezone and it
is necessary to have control over it when loading from pgloader. Hence,
add a timezone option to the IXF option list, that is now explicit and
local to the IXF parser rather than shared with the DBF option list.
After all, it's shared between the CSV command parsing and the Cast
Rules parsing. src/parsers/command-csv.lisp still contains lots of
facilities shared between the file based sources, will need another
series of splits.
Commit 598c860cf5 broke user defined
casting rules by interning "precision" and "scale" in the
pgloader.user-symbols package: those symbols need to be found in the
pgloader.transforms package instead.
Luckily enough the infrastructure to do that was already in place for
cl:nil.
In order to later be able to have more worker threads sharing the
load (multiple readers and/or writers, maybe more specialized threads
too), have all the stats be managed centrally by a single thread. We
already have a "monitor" thread that get passed log messages so that the
output buffer is not subject to race conditions, extend its use to also
deal with statistics messages.
In the current code, we send a message each time we read a row. In some
future commits we should probably reduce the messaging here to something
like one message per batch in the common case.
Also, as a nice side effect of the code simplification and refactoring
this fixes#283 wherein the before/after sections of individual CSV
files within an ARCHIVE command where not counted in the reporting.
