Adam Ierymenko 0e5651f353
1.12.0 merge to main (#2104)
* add note about forceTcpRelay

* Create a sample systemd unit for tcp proxy

* set gitattributes for rust & cargo so hashes dont conflict on Windows

* Revert "set gitattributes for rust & cargo so hashes dont conflict on Windows"

This reverts commit 032dc5c108195f6bbc2e224f00da5b785df4b7f9.

* Turn off autocrlf for rust source

Doesn't appear to play nice well when it comes to git and vendored cargo package hashes

* Fix #1883 (#1886)

Still unknown as to why, but the call to `nc->GetProperties()` can fail
when setting a friendly name on the Windows virtual ethernet adapter.
Ensure that `ncp` is not null before continuing and accessing the device
GUID.

* Don't vendor packages for zeroidc (#1885)

* Added docker environment way to join networks (#1871)

* add StringUtils

* fix headers
use recommended headers and remove unused headers

* move extern "C"
only JNI functions need to be exported

* cleanup

* fix ANDROID-50: RESULT_ERROR_BAD_PARAMETER typo

* fix typo in log message

* fix typos in JNI method signatures

* fix typo

* fix ANDROID-51: fieldName is uninitialized

* fix ANDROID-35: memory leak

* fix missing DeleteLocalRef in loops

* update to use unique error codes

* add GETENV macro

* add LOG_TAG defines

* ANDROID-48: add ZT_jnicache.cpp

* ANDROID-48: use ZT_jnicache.cpp and remove ZT_jnilookup.cpp and ZT_jniarray.cpp

* add Event.fromInt

* add PeerRole.fromInt

* add ResultCode.fromInt

* fix ANDROID-36: issues with ResultCode

* add VirtualNetworkConfigOperation.fromInt

* fix ANDROID-40: VirtualNetworkConfigOperation out-of-sync with ZT_VirtualNetworkConfigOperation enum

* add VirtualNetworkStatus.fromInt

* fix ANDROID-37: VirtualNetworkStatus out-of-sync with ZT_VirtualNetworkStatus enum

* add VirtualNetworkType.fromInt

* make NodeStatus a plain data class

* fix ANDROID-52: synchronization bug with nodeMap

* Node init work: separate Node construction and init

* add Node.toString

* make PeerPhysicalPath a plain data class

* remove unused PeerPhysicalPath.fixed

* add array functions

* make Peer a plain data class

* make Version a plain data class

* fix ANDROID-42: copy/paste error

* fix ANDROID-49: VirtualNetworkConfig.equals is wrong

* reimplement VirtualNetworkConfig.equals

* reimplement VirtualNetworkConfig.compareTo

* add VirtualNetworkConfig.hashCode

* make VirtualNetworkConfig a plain data class

* remove unused VirtualNetworkConfig.enabled

* reimplement VirtualNetworkDNS.equals

* add VirtualNetworkDNS.hashCode

* make VirtualNetworkDNS a plain data class

* reimplement VirtualNetworkRoute.equals

* reimplement VirtualNetworkRoute.compareTo

* reimplement VirtualNetworkRoute.toString

* add VirtualNetworkRoute.hashCode

* make VirtualNetworkRoute a plain data class

* add isSocketAddressEmpty

* add addressPort

* add fromSocketAddressObject

* invert logic in a couple of places and return early

* newInetAddress and newInetSocketAddress work
allow newInetSocketAddress to return NULL if given empty address

* fix ANDROID-38: stack corruption in onSendPacketRequested

* use GETENV macro

* JniRef work
JniRef does not use callbacks struct, so remove
fix NewGlobalRef / DeleteGlobalRef mismatch

* use PRId64 macros

* switch statement work

* comments and logging

* Modifier 'public' is redundant for interface members

* NodeException can be made a checked Exception

* 'NodeException' does not define a 'serialVersionUID' field

* 'finalize()' should not be overridden
this is fine to do because ZeroTierOneService calls close() when it is done

* error handling, error reporting, asserts, logging

* simplify loadLibrary

* rename Node.networks -> Node.networkConfigs

* Windows file permissions fix (#1887)

* Allow macOS interfaces to use multiple IP addresses (#1879)

Co-authored-by: Sean OMeara <someara@users.noreply.github.com>
Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* Fix condition where full HELLOs might not be sent when necessary (#1877)

Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* 1.10.4 version bumps

* Add security policy to repo (#1889)

* [+] add e2k64 arch (#1890)

* temp fix for ANDROID-56: crash inside newNetworkConfig from too many args

* 1.10.4 release notes

* Windows 1.10.4 Advanced Installer bump

* Revert "temp fix for ANDROID-56: crash inside newNetworkConfig from too many args"

This reverts commit dd627cd7f44ad623a110bb14f72d0bea72a09e30.

* actual fix for ANDROID-56: crash inside newNetworkConfig
cast all arguments to varargs functions as good style

* Fix addIp being called with applied ips (#1897)

This was getting called outside of the check for existing ips
Because of the added ifdef and a brace getting moved to the
wrong place.

```
if (! n.tap()->addIp(*ip)) {
	fprintf(stderr, "ERROR: unable to add ip address %s" ZT_EOL_S, ip->toString(ipbuf));
}
WinFWHelper::newICMPRule(*ip, n.config().nwid);

```

* 1.10.5 (#1905)

* 1.10.5 bump

* 1.10.5 for Windows

* 1.10.5

* Prevent path-learning loops (#1914)

* Prevent path-learning loops

* Only allow new overwrite if not bonded

* fix binding temporary ipv6 addresses on macos (#1910)

The check code wasn't running.

I don't know why !defined(TARGET_OS_IOS) would exclude code on
desktop macOS. I did a quick search and changed it to defined(TARGET_OS_MAC).
Not 100% sure what the most correct solution there is.

You can verify the old and new versions with

`ifconfig | grep temporary`

plus

`zerotier-cli info -j` -> listeningOn

* 1.10.6 (#1929)

* 1.10.5 bump

* 1.10.6

* 1.10.6 AIP for Windows.

* Release notes for 1.10.6 (#1931)

* Minor tweak to Synology Docker image script (#1936)

* Change if_def again so ios can build (#1937)

All apple's variables are "defined"
but sometimes they are defined as "0"

* move begin/commit into try/catch block (#1932)

Thread was exiting in some cases

* Bump openssl from 0.10.45 to 0.10.48 in /zeroidc (#1938)

Bumps [openssl](https://github.com/sfackler/rust-openssl) from 0.10.45 to 0.10.48.
- [Release notes](https://github.com/sfackler/rust-openssl/releases)
- [Commits](https://github.com/sfackler/rust-openssl/compare/openssl-v0.10.45...openssl-v0.10.48)

---
updated-dependencies:
- dependency-name: openssl
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* new drone bits

* Fix multiple network join from environment entrypoint.sh.release (#1961)

* _bond_m guards _bond, not _paths_m (#1965)

* Fix: warning: mutex '_aqm_m' is not held on every path through here [-Wthread-safety-analysis] (#1964)

* Bump h2 from 0.3.16 to 0.3.17 in /zeroidc (#1963)

Bumps [h2](https://github.com/hyperium/h2) from 0.3.16 to 0.3.17.
- [Release notes](https://github.com/hyperium/h2/releases)
- [Changelog](https://github.com/hyperium/h2/blob/master/CHANGELOG.md)
- [Commits](https://github.com/hyperium/h2/compare/v0.3.16...v0.3.17)

---
updated-dependencies:
- dependency-name: h2
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* Add note that binutils is required on FreeBSD (#1968)

* Add prometheus metrics for Central controllers (#1969)

* add header-only prometheus lib to ext

* rename folder

* Undo rename directory

* prometheus simpleapi included on mac & linux

* wip

* wire up some controller stats

* Get windows building with prometheus

* bsd build flags for prometheus

* Fix multiple network join from environment entrypoint.sh.release (#1961)

* _bond_m guards _bond, not _paths_m (#1965)

* Fix: warning: mutex '_aqm_m' is not held on every path through here [-Wthread-safety-analysis] (#1964)

* Serve prom metrics from /metrics endpoint

* Add prom metrics for Central controller specific things

* reorganize metric initialization

* testing out a labled gauge on Networks

* increment error counter on throw

* Consolidate metrics definitions

Put all metric definitions into node/Metrics.hpp.  Accessed as needed
from there.

* Revert "testing out a labled gauge on Networks"

This reverts commit 499ed6d95e11452019cdf48e32ed4cd878c2705b.

* still blows up but adding to the record for completeness right now

* Fix runtime issues with metrics

* Add metrics files to visual studio project

* Missed an "extern"

* add copyright headers to new files

* Add metrics for sent/received bytes (total)

* put /metrics endpoint behind auth

* sendto returns int on Win32

---------

Co-authored-by: Leonardo Amaral <leleobhz@users.noreply.github.com>
Co-authored-by: Brenton Bostick <bostick@gmail.com>

* Central startup update (#1973)

* allow specifying authtoken in central startup

* set allowManagedFrom

* move redis_mem_notification to the correct place

* add node checkins metric

* wire up min/max connection pool size metrics

* x86_64-unknown-linux-gnu on ubuntu runner (#1975)

* adding incoming zt packet type metrics (#1976)

* use cpp-httplib for HTTP control plane (#1979)

refactored the old control plane code to use [cpp-httplib](https://github.com/yhirose/cpp-httplib) instead of a hand rolled HTTP server.  Makes the control plane code much more legible.  Also no longer randomly stops responding.

* Outgoing Packet Metrics (#1980)

add tx/rx labels to packet counters and add metrics for outgoing packets

* Add short-term validation test workflow (#1974)

Add short-term validation test workflow

* Brenton/curly braces (#1971)

* fix formatting

* properly adjust various lines
breakup multiple statements onto multiple lines

* insert {} around if, for, etc.

* Fix rust dependency caching (#1983)

* fun with rust caching

* kick

* comment out invalid yaml keys for now

* Caching should now work

* re-add/rename key directives

* bump

* bump

* bump

* Don't force rebuild on Windows build GH Action (#1985)

Switching `/t:ZeroTierOne:Rebuild` to just `/t:ZeroTierOne` allows the Windows build to use the rust cache.  `/t:ZeroTierOne:Rebuild` cleared the cache before building.

* More packet metrics (#1982)

* found path negotation sends that weren't accounted for

* Fix histogram so it will actually compile

* Found more places for packet metrics

* separate the bind & listen calls on the http backplane (#1988)

* fix memory leak (#1992)

* fix a couple of metrics (#1989)

* More aggressive CLI spamming (#1993)

* fix type signatures (#1991)

* Network-metrics (#1994)

* Add a couple quick functions for converting a uint64_t network ID/node ID into std::string

* Network metrics

* Peer metrics (#1995)

* Adding peer metrics

still need to be wired up for use

* per peer packet metrics

* Fix crash from bad instantiation of histogram

* separate alive & dead path counts

* Add peer metric update block

* add peer latency values in doPingAndKeepalive

* prevent deadlock

* peer latency histogram actually works now

* cleanup

* capture counts of packets to specific peers

---------

Co-authored-by: Joseph Henry <joseph.henry@zerotier.com>

* Metrics consolidation (#1997)

* Rename zt_packet_incoming -> zt_packet

Also consolidate zt_peer_packets into a single metric with tx and rx labels.  Same for ztc_tcp_data and ztc_udp_data

* Further collapse tcp & udp into metric labels for zt_data

* Fix zt_data metric description

* zt_peer_packets description fix

* Consolidate incoming/outgoing network packets to a single metric

* zt_incoming_packet_error -> zt_packet_error

* Disable peer metrics for central controllers

Can change in the future if needed, but given the traffic our controllers serve, that's going to be a *lot* of data

* Disable peer metrics for controllers pt 2

* Update readme files for metrics (#2000)

* Controller Metrics & Network Config Request Fix (#2003)

* add new metrics for network config request queue size and sso expirations
* move sso expiration to its own thread in the controller
* fix potential undefined behavior when modifying a set

* Enable RTTI in Windows build

The new prometheus histogram stuff needs it.

Access violation - no RTTI data!INVALID packet 636ebd9ee8cac6c0 from cafe9efeb9(2605:9880:200:1200:30:571:e34:51/9993) (unexpected exception in tryDecode())

* Don't re-apply routes on BSD

See issue #1986

* Capture setContent by-value instead of by-reference (#2006)

Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* fix typos (#2010)

* central controller metrics & request path updates (#2012)

* internal db metrics

* use shared mutexes for read/write locks

* remove this lock. only used for a metric

* more metrics

* remove exploratory metrics

place controller request benchmarks behind ifdef

* Improve validation test (#2013)

* fix init order for EmbeddedNetworkController (#2014)

* add constant for getifaddrs cache time

* cache getifaddrs - mac

* cache getifaddrs - linux

* cache getifaddrs - bsd

* cache getifaddrs - windows

* Fix oidc client lookup query

join condition referenced the wrong table.  Worked fine unless there were multiple identical client IDs

* Fix udp sent metric

was only incrementing by 1 for each packet sent

* Allow sending all surface addresses to peer in low-bandwidth mode

* allow enabling of low bandwidth mode on controllers

* don't unborrow bad connections

pool will clean them up later

* Multi-arch controller container (#2037)

create arm64 & amd64 images for central controller

* Update README.md

issue #2009

* docker tags change

* fix oidc auth url memory leak (#2031)

getAuthURL() was not calling zeroidc::free_cstr(url);

the only place authAuthURL is called, the url can be retrieved
from the network config instead.

You could alternatively copy the string and call free_cstr in getAuthURL.
If that's better we can change the PR.

Since now there are no callers of getAuthURL I deleted it.

Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* Bump openssl from 0.10.48 to 0.10.55 in /zeroidc (#2034)

Bumps [openssl](https://github.com/sfackler/rust-openssl) from 0.10.48 to 0.10.55.
- [Release notes](https://github.com/sfackler/rust-openssl/releases)
- [Commits](https://github.com/sfackler/rust-openssl/compare/openssl-v0.10.48...openssl-v0.10.55)

---
updated-dependencies:
- dependency-name: openssl
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* zeroidc cargo warnings (#2029)

* fix unused struct member cargo warning

* fix unused import cargo warning

* fix unused return value cargo warning

---------

Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* fix memory leak in macos ipv6/dns helper (#2030)

Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>

* Consider ZEROTIER_JOIN_NETWORKS in healthcheck (#1978)

* Add a 2nd auth token only for access to /metrics (#2043)

* Add a 2nd auth token for /metrics

Allows administrators to distribute a token that only has access to read
metrics and nothing else.

Also added support for using bearer auth tokens for both types of tokens

Separate endpoint for metrics #2041

* Update readme

* fix a couple of cases of writing the wrong token

* Add warning to cli for allow default on FreeBSD

It doesn't work.
Not possible to fix with deficient network
stack and APIs.

ZeroTierOne-freebsd # zerotier-cli set 9bee8941b5xxxxxx allowDefault=1
400 set Allow Default does not work properly on FreeBSD. See #580
root@freebsd13-a:~/ZeroTierOne-freebsd # zerotier-cli get 9bee8941b5xxxxxx allowDefault
1

* ARM64 Support for TapDriver6 (#1949)

* Release memory previously allocated by UPNP_GetValidIGD

* Fix ifdef that breaks libzt on iOS (#2050)

* less drone (#2060)

* Exit if loading an invalid identity from disk (#2058)

* Exit if loading an invalid identity from disk

Previously, if an invalid identity was loaded from disk, ZeroTier would
generate a new identity & chug along and generate a brand new identity
as if nothing happened.  When running in containers, this introduces the
possibility for key matter loss; especially when running in containers
where the identity files are mounted in the container read only.  In
this case, ZT will continue chugging along with a brand new identity
with no possibility of recovering the private key.

ZeroTier should exit upon loading of invalid identity.public/identity.secret #2056

* add validation test for #2056

* tcp-proxy: fix build

* Adjust tcp-proxy makefile to support metrics

There's no way to get the metrics yet. Someone will
have to add the http service.

* remove ZT_NO_METRIC ifdef

* Implement recvmmsg() for Linux to reduce syscalls. (#2046)

Between 5% and 40% speed improvement on Linux, depending on system configuration and load.

* suppress warnings: comparison of integers of different signs: 'int64_t' (aka 'long') and 'uint64_t' (aka 'unsigned long') [-Wsign-compare] (#2063)

* fix warning: 'OS_STRING' macro redefined [-Wmacro-redefined] (#2064)

Even though this is in ext, these particular chunks of code were added
by us, so are ok to modify.

* Apply default route a different way - macOS

The original way we applied default route, by forking
0.0.0.0/0 into 0/1 and 128/1 works, but if mac os has any networking
hiccups -if you change SSIDs or sleep/wake- macos erases the system default route.
And then all networking on the computer is broken.

to summarize the new way:
allowDefault=1
```
sudo route delete default 192.168.82.1
sudo route add default 10.2.0.2
sudo route add -ifscope en1 default 192.168.82.1
```

gives us this routing table
```
Destination        Gateway            RT_IFA             Flags        Refs      Use    Mtu          Netif Expire    rtt(ms) rttvar(ms)
default            10.2.0.2           10.2.0.18          UGScg          90        1   2800       feth4823
default            192.168.82.1       192.168.82.217     UGScIg
```

allowDefault=0
```
sudo route delete default
sudo route delete -ifscope en1 default
sudo route add default 192.168.82.1
```

Notice the I flag, for -ifscope, on the physical default route.

route change does not seem to work reliably.

* fix docker tag for controllers (#2066)

* Update build.sh (#2068)

fix mkwork compilation errors

* Fix network DNS on macOS

It stopped working for ipv4 only networks in Monterey.
See #1696

We add some config like so to System Configuration

```
scutil
show State:/Network/Service/9bee8941b5xxxxxx/IPv4
<dictionary> {
  Addresses : <array> {
    0 : 10.2.1.36
  }
  InterfaceName : feth4823
  Router : 10.2.1.36
  ServerAddress : 127.0.0.1
}

```

* Add search domain to macos dns configuration

Stumbled upon this while debugging something else.
If we add search domain to our system configuration for
network DNS, then search domains work:

```
ping server1                                                                                                                                                                                    ~
PING server1.my.domain (10.123.3.1): 56 data bytes
64 bytes from 10.123.3.1
```

* Fix reporting of secondaryPort and tertiaryPort See: #2039

* Fix typos (#2075)

* Disable executable stacks on assembly objects (#2071)

Add `--noexecstack` to the assembler flags so the resulting binary
will link with a non-executable stack.

Fixes zerotier/ZeroTierOne#1179

Co-authored-by: Joseph Henry <joseph.henry@zerotier.com>

* Test that starting zerotier before internet works

* Don't skip hellos when there are no paths available

working on #2082

* Update validate-1m-linux.sh

* Save zt node log files on abort

* Separate test and summary step in validator script

* Don't apply default route until zerotier is "online"

I was running into issues with restarting the zerotier service while
"full tunnel" mode is enabled.
When zerotier first boots, it gets network state from the cache
on disk. So it immediately applies all the routes it knew about
before it shutdown.
The network config may have change in this time.
If it has, then your default route is via a route
you are blocked from talking on. So you  can't get the current
network config, so your internet does not work.

Other options include
- don't use cached network state on boot
- find a better criteria than "online"

* Fix node time-to-online counter in validator script

* Export variables so that they are accessible by exit function

* Fix PortMapper issue on ZeroTier startup

See issue #2082

We use a call to libnatpmp::ininatpp to make sure the computer
has working network sockets before we go into the main
nat-pmp/upnp logic.

With basic exponenetial delay up to 30 seconds.

* testing

* Comment out PortMapper debug

this got left turned on in a confusing merge previously

* fix macos default route again

see commit fb6af1971 * Fix network DNS on macOS
adding that stuff to System Config causes this extra route to be added
which breaks ipv4 default route.
We figured out a weird System Coniguration setting
that works.

--- old
couldn't figure out how to fix it in SystemConfiguration
so here we are# Please enter the commit message for your changes. Lines starting

We also moved the dns setter to before the syncIps stuff
to help with a race condition. It didn't always work when
you re-joined a network with default route enabled.

* Catch all conditions in switch statement, remove trailing whitespaces

* Add setmtu command, fix bond lifetime issue

* Basic cleanups

* Check if null is passed to VirtualNetworkConfig.equals and name fixes

* ANDROID-96: Simplify and use return code from node_init directly

* Windows arm64 (#2099)

* ARM64 changes for 1.12

* 1.12 Windows advanced installer updates and updates for ARM64

* 1.12.0

* Linux build fixes for old distros.

* release notes

---------

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: travis laduke <travisladuke@gmail.com>
Co-authored-by: Grant Limberg <grant.limberg@zerotier.com>
Co-authored-by: Grant Limberg <glimberg@users.noreply.github.com>
Co-authored-by: Leonardo Amaral <leleobhz@users.noreply.github.com>
Co-authored-by: Brenton Bostick <bostick@gmail.com>
Co-authored-by: Sean OMeara <someara@users.noreply.github.com>
Co-authored-by: Joseph Henry <joseph-henry@users.noreply.github.com>
Co-authored-by: Roman Peshkichev <roman.peshkichev@gmail.com>
Co-authored-by: Joseph Henry <joseph.henry@zerotier.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Stavros Kois <47820033+stavros-k@users.noreply.github.com>
Co-authored-by: Jake Vis <jakevis@outlook.com>
Co-authored-by: Jörg Thalheim <joerg@thalheim.io>
Co-authored-by: lison <imlison@foxmail.com>
Co-authored-by: Kenny MacDermid <kenny@macdermid.ca>
2023-08-23 14:24:21 -04:00

238 lines
7.7 KiB
C++

/* Definition of the pqxx::pipeline class.
*
* Throughput-optimized mechanism for executing queries.
*
* DO NOT INCLUDE THIS FILE DIRECTLY; include pqxx/pipeline instead.
*
* Copyright (c) 2000-2022, Jeroen T. Vermeulen.
*
* See COPYING for copyright license. If you did not receive a file called
* COPYING with this source code, please notify the distributor of this
* mistake, or contact the author.
*/
#ifndef PQXX_H_PIPELINE
#define PQXX_H_PIPELINE
#if !defined(PQXX_HEADER_PRE)
# error "Include libpqxx headers as <pqxx/header>, not <pqxx/header.hxx>."
#endif
#include <limits>
#include <map>
#include <string>
#include "pqxx/transaction_base.hxx"
namespace pqxx
{
// TODO: libpq 14 introduced a similar "pipeline mode." Can we use that?
/// Processes several queries in FIFO manner, optimized for high throughput.
/** Use a pipeline if you want to keep doing useful work while your queries are
* executing. Result retrieval is decoupled from execution request; queries
* "go in at the front" and results "come out the back."
*
* Actually, you can retrieve the results in any order if you want, but it may
* lead to surprising "time travel" effects if any of the queries fails. In
* particular, syntax errors in the queries can confuse things and show up too
* early in the stream of results.
*
* Generally, if any of the queries fails, it will throw an exception at the
* point where you request its result. But it may happen earlier, especially
* if you request results out of chronological order.
*
* @warning While a pipeline is active, you cannot execute queries, open
* streams, etc. on the same transaction. A transaction can have at most one
* object of a type derived from @ref pqxx::transaction_focus active on it at a
* time.
*/
class PQXX_LIBEXPORT pipeline : public transaction_focus
{
public:
/// Identifying numbers for queries.
using query_id = long;
pipeline(pipeline const &) = delete;
pipeline &operator=(pipeline const &) = delete;
/// Start a pipeline.
explicit pipeline(transaction_base &t) : transaction_focus{t, s_classname}
{
init();
}
/// Start a pipeline. Assign it a name, for more helpful error messages.
pipeline(transaction_base &t, std::string_view tname) :
transaction_focus{t, s_classname, tname}
{
init();
}
/// Close the pipeline.
~pipeline() noexcept;
/// Add query to the pipeline.
/** Queries accumulate in the pipeline, which sends them to the backend in a
* batch separated by semicolons. The queries you insert must not use this
* trick themselves, or the pipeline will get hopelessly confused!
*
* @return Identifier for this query, unique only within this pipeline.
*/
query_id insert(std::string_view) &;
/// Wait for all ongoing or pending operations to complete, and detach.
/** Detaches from the transaction when done.
*
* This does not produce the queries' results, so it may not report any
* errors which may have occurred in their execution. To be sure that your
* statements succeeded, call @ref retrieve until the pipeline is empty.
*/
void complete();
/// Forget all ongoing or pending operations and retrieved results.
/** Queries already sent to the backend may still be completed, depending
* on implementation and timing.
*
* Any error state (unless caused by an internal error) will also be cleared.
* This is mostly useful in a nontransaction, since a backend transaction is
* aborted automatically when an error occurs.
*
* Detaches from the transaction when done.
*/
void flush();
/// Cancel ongoing query, if any.
/** May cancel any or all of the queries that have been inserted at this
* point whose results have not yet been retrieved. If the pipeline lives in
* a backend transaction, that transaction may be left in a nonfunctional
* state in which it can only be aborted.
*
* Therefore, either use this function in a nontransaction, or abort the
* transaction after calling it.
*/
void cancel();
/// Is result for given query available?
[[nodiscard]] bool is_finished(query_id) const;
/// Retrieve result for given query.
/** If the query failed for whatever reason, this will throw an exception.
* The function will block if the query has not finished yet.
* @warning If results are retrieved out-of-order, i.e. in a different order
* than the one in which their queries were inserted, errors may "propagate"
* to subsequent queries.
*/
result retrieve(query_id qid)
{
return retrieve(m_queries.find(qid)).second;
}
/// Retrieve oldest unretrieved result (possibly wait for one).
/** @return The query's identifier and its result set. */
std::pair<query_id, result> retrieve();
[[nodiscard]] bool empty() const noexcept { return std::empty(m_queries); }
/// Set maximum number of queries to retain before issuing them to the
/// backend.
/** The pipeline will perform better if multiple queries are issued at once,
* but retaining queries until the results are needed (as opposed to issuing
* them to the backend immediately) may negate any performance benefits the
* pipeline can offer.
*
* Recommended practice is to set this value no higher than the number of
* queries you intend to insert at a time.
* @param retain_max A nonnegative "retention capacity;" passing zero will
* cause queries to be issued immediately
* @return Old retention capacity
*/
int retain(int retain_max = 2) &;
/// Resume retained query emission. Harmless when not needed.
void resume() &;
private:
struct PQXX_PRIVATE Query
{
explicit Query(std::string_view q) :
query{std::make_shared<std::string>(q)}
{}
std::shared_ptr<std::string> query;
result res;
};
using QueryMap = std::map<query_id, Query>;
void init();
void attach();
void detach();
/// Upper bound to query id's.
static constexpr query_id qid_limit() noexcept
{
// Parenthesise this to work around an eternal Visual C++ problem:
// Without the extra parentheses, unless NOMINMAX is defined, the
// preprocessor will mistake this "max" for its annoying built-in macro
// of the same name.
return (std::numeric_limits<query_id>::max)();
}
/// Create new query_id.
PQXX_PRIVATE query_id generate_id();
bool have_pending() const noexcept
{
return m_issuedrange.second != m_issuedrange.first;
}
PQXX_PRIVATE void issue();
/// The given query failed; never issue anything beyond that.
void set_error_at(query_id qid) noexcept
{
PQXX_UNLIKELY
if (qid < m_error)
m_error = qid;
}
/// Throw pqxx::internal_error.
[[noreturn]] PQXX_PRIVATE void internal_error(std::string const &err);
PQXX_PRIVATE bool obtain_result(bool expect_none = false);
PQXX_PRIVATE void obtain_dummy();
PQXX_PRIVATE void get_further_available_results();
PQXX_PRIVATE void check_end_results();
/// Receive any results that happen to be available; it's not urgent.
PQXX_PRIVATE void receive_if_available();
/// Receive results, up to stop if possible.
PQXX_PRIVATE void receive(pipeline::QueryMap::const_iterator stop);
std::pair<pipeline::query_id, result> retrieve(pipeline::QueryMap::iterator);
QueryMap m_queries;
std::pair<QueryMap::iterator, QueryMap::iterator> m_issuedrange;
int m_retain = 0;
int m_num_waiting = 0;
query_id m_q_id = 0;
/// Is there a "dummy query" pending?
bool m_dummy_pending = false;
/// Point at which an error occurred; no results beyond it will be available
query_id m_error = qid_limit();
/// Encoding.
/** We store this in the object to avoid the risk of exceptions at awkward
* moments.
*/
internal::encoding_group m_encoding;
static constexpr std::string_view s_classname{"pipeline"};
};
} // namespace pqxx
#endif