I’ve been having some landscaping work done in my yard, and despite having all of the utility lines marked, the crews have managed to cut my cable internet line on several occasions. Since I work from home, and we don’t have hardly any cellular reception at the house, it was pretty devastating. I couldn’t even call the cable company to repair it without driving somewhere with better reception.
So when they said it would be a few days to get a technician to come and repair the cut cable, I decided to try out some repairs on my own. I happen to have some coax crimpers, spare cable, and ends at home, so I first attempted to terminate the cut ends, but the direct-burial cable that is used outdoors is quite a bit thicker than indoor cable, so my ends and crimpers wouldn’t work.
Without any of the correct tools, I was left with just the most primitive of methods: simply twist the center wire together with some needle nose plier, and tie them together with wire ties.
Here’s one of my first attempts when they cut the coax. I tied it into my own coax and which ran back to the house. On this first attempt, I tried to leave some extra shielding and twist that together from each end.
A subsequent cut, with newer cable had enough slack that I could just tie the two ends together directly. After it was repaired, this is what the technician left so you can see how I first twisted the ends together as much as I could with some pliers before adding a wire tie onto it. I just cut the shielding clean off and didn’t attempt to mess with it, which still resulted in it working fine.
It didn’t result in the full 200 Mbps+ speed that I should be getting, but 50+ Mbps was absolutely better than nothing for the few days until the technician could come and re-terminate the ends properly:
I was surprised that the Ubiquity Dream Machine Pro doesn’t have SNMP available. I recall that there was an option to enable it in older versions of their software, but the current 3.0.20 version doesn’t even have an option to enable it (and I don’t think that it worked correctly in previous versions).
Fortunately, its basically just a Debian machine, so you can enable it yourself! These are the steps that I took to enable snmpd so that I could add it to my network monitoring system:
First, update the respositories and install the snmp and snmpd packages:
apt update
apt install -y snmp snmpd
Then, you have to edit the snmpd.conf file in /etc/snmp/snmpd.conf and change these two lines from the View section. This change makes it so that instead of providing information only about the host system, it provides information about all of the attached interfaces as well.
view systemonly included .1.3.6.1.2.1.1
view systemonly included .1.3.6.1.2.1.25.1
To these two lines (note you remove the final .1 from the end of each).
view systemonly included .1.3.6.1.2.1
view systemonly included .1.3.6.1.2.1.25.1.1
Also, you’ll probably want to configure the snmpd deamon so that it will be available on a local network interface, so change the agentaddress line to this (obviously, with your box’s IP address if it isn’t 192.168.0.1):
agentaddress 127.0.0.1,[::1],192.168.0.1
Then restart the snmpd deamon
service snmpd restart
You can test that it is working by running snmpwalk with a command like this:
snmpwalk -Os -c public -v 2c 192.168.0.1
Which should output hundreds of lines of stuff that start out similar to this:
If that works, congratulations! You’ve got snmpd installed on your Ubiquity Dream Machine Pro. Your network monitoring system may take a little time for it to notice that SNMP statistics are now available on the device.
Note that upgrading the device will probably lose these configs and they’d have to be re-done.
In one of my applications, I’ve been noticing this error occurring more frequently.
PHP Warning: Packets out of order. Expected 1 received 0. Packet size=145
When investigating, I ran this long running command in the foreground and watched for a pattern. Sure enough, I found that when the program waited a long time between jobs, that the first command when it resumed would result in this error.
My application had some retry logic built-in, so that it resumed and went on as normal, so it was just an annoyance, but I don’t like it when I don’t understand how things are working.
I was able to recreate this problem reliably with this short script:
<?php
require_once 'include.php'; // Connects to the database
// Set the session wait_timeout to a small value
$db->query("SET session wait_timeout=10;");
// Prove that the connection works
$one = $db->getOne("SELECT 1");
echo "Got one = {$one}\n";
// Sleep for longer than the wait_timeout
sleep(11)
// Retry the query
$one = $db->getOne("SELECT 1");
echo "Got one = {$one}\n";
When executed, it provided this output, concluding that the wait_timeout is the problem:
got one = 1
PHP Warning: Packets out of order. Expected 1 received 0. Packet size=145 in /path/to/myapp/db.class.php on line 68
PHP Stack trace:
PHP 1. {main}() /path/to/myapp/dbtest.php:0
PHP 2. db->getOne($sql = 'SELECT 1', $args = *uninitialized*, $recurse = *uninitialized*) /path/to/myapp/dbtest.php:13
PHP 3. PDOStatement->execute($params = []) /path/to/myapp/db.class.php:68
To prevent this problem, I implemented a timer that counts the time between queries and reconnects to the server if wait_timeout seconds elapses between queries. This may not be exact, because it counts the time between the start of the query, but it largely prevented this problem.
In my database connection class (db.class.php), it calls the conn() method for each query, so I added the timer here which causes it to disconnect when there is more than $sqlTimeout seconds between SQL queries
class db
{
protected $lastActivityTs = null;
static protected $sqlTimeout = 3600; // Make sure you copy this value from your MySQL Server
public function conn()
{
if (isset($this->dbh) && (microtime(true) - $this->lastActivityTs) >= self::$sqlTimeout) {
echo "Disconnecting after expired SQL connection\n";
// Our connection is probably timed out by the server anyway
$this->disconnect();
}
if (!isset($this->dbh)) {
$this->_connect();
}
$this->lastActivityTs = microtime(true);
return $this->dbh;
}
Note that our library here automatically retries once when a connection error occurs. This has also been important to catch temporary failures and disconnects from the MySQL server and have it retry the connection.
// Continuing in the db class
public function getOne($sql, $args = [], $recurse = true)
{
try {
$sth = $this->conn()->prepare($sql);
$sth->execute($args);
$sth->setFetchMode(PDO::FETCH_NUM);
$row = $sth->fetch();
return $row[0] ?? null;
} catch (PDOException $e) {
if ($recurse && 'HY000' == $e->getCode()) {
// SQLSTATE[HY000]: General error: 2013 Lost connection to MySQL server during query
unset($this->dbh);
return $this->getOne($sql, $args, false);
}
throw $e;
}
}
We have one database server that is running on an old version of Aurora based on MySQL 5.6. AWS is deprecating that version soon and it needs to be upgraded, so I have been working on replacing it. Upgrading the existing 5.6 server to 5.7, then to 8.0 isn’t an option due to an impossibly huge InnoDB transaction history list that will never fix itself. Plus, I want to improve a couple of other things along the way.
I made several attempts and migrating from Aurora 5.6 to Aurora 8.0, but during that process, I grew tired of Aurora quirks and costs. Here are some of my raw notes on what was an embarrassingly long migration of a database server from Aurora to MySQL. Going from MySQL to Aurora took just a couple of clicks. But converting from Aurora back to MySQL took months and a lot of headaches.
TLDR: Along the way, I tried Using Amazon’s Database Migration Service, but eventually gave up for a good old closely monitored mysqldump and custom scripts.
I had a few goals/requirements:
Get rid of or soon-to-be-deprecated Aurora instance based on MySQL 5.6
Stop Paying for Storage IOPS (often over $100/day)
Convert tables from utf8mb3 to utf8mb4
Minimal downtime or customer disruption. Some disruption during low-usage times is okay.
A new MySQL 8 instance with a GP3 storage volume and the recently announced RDS Optimized Writes means that MySQL should be able to handle the workload with no problem, and gets this server back into the MySQL realm, where all of our other servers are, and with which we are more comfortable.
Attempts at using AWS Database Migration Service (DMS)
This service looked promising, but has a learning curve. I eventually gave up using it because of repeated problems that would have taken too much effort to try and resolve.
First attempts:
On the surface, it seems like you configure a source, configure a destination, and then tell DMS to sync one to the other and keep them in sync. It does this in two Phases: the Full Dump, and the Change Data Capture (CDC). I learned the hard way that the Full Dump doesn’t include any indexes on the tables! This is done to make it as fast as possible. The second, CDC Phase, just executes statements from the binary log, so without indexes on a 400+G table, they take forever and this will never work.
I also concluded that one of our 300+GB tables can actually be done in a separate process, after the rest of the data is loaded. It contains historic information that will make some things in the application look incomplete until it is loaded, but the application will work with it empty.
Second attempts:
Used DMS for the full dump, the configured it to stop after the full dump, before starting the CDC Process. While it is stopped, I added the database indexes and foreign keys. I tried this several times with varying degrees of success and trying to minimize the amount of time that it took to add the indexes. Some tables were done instantly, some took a couple hours, and some were 12+ hours. At one point I had figured it would take about 62 hours to add the indexes. I think I got that down to 39 hours by increasing the IOPS, running some ALTER TABLES in parallel, etc.
After indexes were added, I started the second phase of DMS – the Change Data Capture is supposed to pick up in time where the Full Dump was taken, and then apply all of the changes from the Binary Logs to the new server. That process didn’t go smoothly. Again, the first attempts looked promising, but then the binary logs on the server were deleted, so it couldn’t continue. I increased the number of days that binary logs were kept, and made more attempts, but they had problems with foreign key and unique constraints on tables.
The biggest problem with these attempts was that it took about 24 hours for the data migration, and about 48 hours to add indexes. So each attempt was several days effort.
Third and last attempts at using DMS:
After getting pretty familiar DMS, I ended up creating the schema via `mysqldump –no-data` then manually editing the file to exclude indexes on some of the biggest tables that would cause the import to go slow. I excluded the one large, historic table. My overall process looked like this:
Edit /tmp/schema-limited-indexes.sql and remove foreign keys and indexes on large tables
cat /tmp/schema-limited-indexes.sql | mysql –defaults-group-suffix=newserver thedatabase
On the new server, run ALTER TABLE the_historic_table ENGINE=blackhole;
Start DMS process, make sure to have it stop between Full Load and CDC.
Wait ~24+ hours for Full load to complete
Add Indexes back that were removed from the schema. I had a list of ALTER TABLE statements to run, with an estimate time that each should take. That was estimated at 39 hours
Start second Phase (CDC) of the DMS Task
Wait for CDC to complete (time estimate unknown. The faster the above steps worked, the less it had to replay)
Unfortunately, a couple of attempts at this had the CDC phase still fail with Foreign key constraints. I tried several times and don’t know why this happened. Finding the offending rows took many hours since the queries didn’t have indexes and had to do full table scans. In some cases, there were just a few, to a few-dozen rows that existed in one table without the corresponding row in the foreign table. Its as if the binary log position taken when the snapshot was started was off by a few seconds and the dumps of different tables were started at slightly different positions.
After several attempts (taking a couple weeks), I finally gave up on the DMS approach.
Using MySQL Dump
Using mysqldump to move data from one database server to another is a process I have done thousands of times and written many scripts around. It is pretty well understood and predictable. I did a few trial runs to put together this process:
Temporarily Stop all processes on the master server
Stop all background processes that write to the server
Change the password so that no processes can write to the master
Execute SHOW BINARY LOGS on master and note the last binary log file and position. Do this a few times to make sure that it does not change. (Note that this would be easier if RDS allowed FLUSH TABLES WITH READ LOCK, but since it doesn’t, this process should work.
Dump the schema to the new server
This has the sed commands in the middle to convert the old “utf8” colations to the desired “utf8mb4” versions. When dumping 1TB+ of data, I found it helped performance a bit to do the schema changes with the sed commands first. That way the bulk of the data doesn’t have to go through these two commands.
mysqldump --defaults-group-suffix=dumpschema --no-data thedatabase |sed "s/utf8 /utf8mb4 /" | sed "s/utf8_/utf8mb4_/" | mysql thedatabase
.my.cnf contains this section with the relevant parameters for the dump
Note that the above command includes the linux pv in between which is a nice way to monitor the progress. It displays a simple line to stderr that allows you to see the total transfer size, elapsed time, and current speed.
266.5GiB 57:16:47 [ 100KiB/s] [ <=> ]
I experimented with several values for the NET_BUFFER_LENGTH parameter by dumping the same multi-GB table over and over with different values for NET_BUFFER_LENGTH. The size of this value determines how many values are included in the INSERT INTO statement generated by mysqldump. I was hoping that a larger value would improve performance, but I found that larger values slowed down. I found the best value was to use 256k.
NET_BUFFER_LENGTH value
Elapsed Time
64k
13m 44s
256k
8m 27s
256k
7m 20s
1M
10m 23s
16M
11m 32s
After Migration is Started
After the mysqldump has been started, I re-enabled traffic back to the master server by setting the password back to the original. I kept all background jobs disabled to minimize the amount of data that had to be copied over afterwards.
Final attempt to use DMS
After the mysqldump was finished, I attempted to use the DMS Change Data Capture process to copy over the data that had changed on the master. You can start a Database Migration Task that begins at a specific point in the Master Log position. Maybe. I tried, it, but it failed pretty quickly with a duplicate key constraint. I gave up on DMS and figured I would just move over any data needed manually via custom scripts.
Other findings
In attempting to maximimize the speed of the transfer, I attempted to increase the IOPS on the GP3 volume from its base level of 12,000 to 32,000. Initially that helped, but for some reason I still don’t understand, the throughput was then limited very strictly to 6,000 IOPS. As seen in the chart below, it bursted above that for some short parts, but it was pretty strictly constrained for most of the time. I think this has to do with how RDS uses multiple volumes to store the data. I suspect that each volume has 6,000 capacity, and all of my data was going to a single volume.
RDS IOPS Maxed at 6,000
That concludes the notes that I wanted to take. Hopefully somebody else finds these learnings or settings useful. If this has been helpful, or if you have any comments on some of the problems that I experienced, please let me know in the comments below.
5+ years ago one of my companies launched a product that is effectively a search engine monitoring tool. Is saves a lot of information about search engine results and the destination pages, then allows the users to see for which search phrases each pages ranks.
The workload is heavily write intensive. No matter the number of users we have to perform a bunch of data collection and save that into our database. A large increase in the number of users would increase the amount of reads, but the base workload of collecting all of the results remains the dominant workload for the database server.
We built this originally using MySQL 5.6, which we had used and managed extensively. We began having concerns with write capacity about the time the that AWS was starting to push Aurora as an alternative to MySQL, with cost and performance benefits. It seemed like an easy win, so we clicked the couple buttons and within minutes our database server was converted from MySQL to Aurora.
Things worked well for a long time. The product worked well and customers liked it. We tweaked the application here and there, but most of the base functionality just continued to do its thing. We moved on to developing other products and maintaining this one.
Fast forward a few years and we found that minor complaints had started to pile up. We add some indexes, make some code and queries more efficient. Adding indexes or altering a 500Gb table has some challenges, but there are tools like pt-online-schema-change that make table changes a little easier without downtime.
As time went on, we got better about allocating costs to the each product that we run and I did start to notice that the cost to run the Aurora instance was quite high. The instance cost itself was predictable, but the pricing of Aurora Database Storage includes a seemingly small cost of $0.20 per million I/O requests that was adding up to sometimes $200+ per day! It was at this point that I started to call Aurora a “Pay for Performance” product. Because it had the ability to scale I/O capacity quite high, inefficient queries executed fast enough not to notice. You just get charged more for them! It can be difficult to track down inefficient queries when everything is running fast. Performance Insights was very helpful to track down queries that could be optimized. By adding some indexes, we reduced our Database I/O and got our I/O costs down to under $100/day. On a traditional MySQL instance, with more limited I/O Capacity, these queries would have been more obvious, as they would have executed more slowly and our traditional troubleshooting would have brought them to our attention for the same optimizations. The “pay for performance” aspect of Aurora kept us from fixing the inefficient queries because they were hidden by being charged more.
Comparing Aurora Billed IO’s to MySQL IOPS
In November 2022 AWS announced that GP3 volumes are now available for RDS instances. The public documentation mentions a 3,000 IPS base capacity but doesn’t mention that for 400G+ volumes, that AWS actually spreads your storage over four volumes for theoretical base 12,000 IOPS. For an additional $0.02/IOPS you can increase your capacity up to 64,000 IOPS. So on the high end, the extra 52,000 extra IOPS at $0.02 comes to $1,040/month or about $35/day. There may be additional throughput needed as well, but for our workload, I found that IOPS was the bottleneck more than throughput.
Since we were still paying $60-$100 most days for Aurora Storage IOPS, it makes sense cost-wise to switch back from Aurora to MySQL. I also favor MySQL because it’s what we’re already used to. I’ve always thought that the monitoring and metrics available on Aurora instances wasn’t up to par with the MySQL instances. And there is just enough of a “black box” in Aurora that it makes things difficult.
In trying to estimate how much IOPS we needed if we switch back to MySQL, I found it a bit of work to estimate how much Aurora was using in terms that I’m used to seeing for MySQL. The only IO metrics available are “[Billed] Volume Read IOPS” and “[Billed] Volume Write IOPS”. These are under the “Cluster” in Cloudwatch Metrics and look like they are billed at 1-hour granularities. Make sure to use the “Sum” statistic instead of “Average” or else you will be off a lot! My server had values values of around 4,000,000 to 13,000,000 for reads and 5,000,000 to 15,000,000 for writes. These values lined up pretty well to costs per day that I was able to see in Cost Explorer. When Cloudwatch Metrics showed a combined 500M IO’s for a day, I was charged $100. To convert the “Billed IOs” that Aurora reports, you have to divide by the number of seconds in the period. If looking at one-hour period, the 9,000,000 IO’s averages out to 2,500 IOPS (divide by 3600 seconds). 30,000,000 IO’s in an hour equates to an average of 8,333 IOPS.
Note that these are averages over an entire hour, so peaks within that hour could be dramatically higher! This gave me confidence that the 12,000 baseline IOPS and availability to pay for up to 64,000 IOPS with GP3 volumes should be able to perform the same workload that was being handled by Aurora.
The effect of Double-Writes
Also, announced in the past month was support for RDS Optimized Writes on newly launched instances within certain instance types. Its unclear if Aurora already has this type of feature enabled, so I’m not certain if the Billed Aurora IO Writes mentioned above would be the number calculated from there, or potentially half of that. Please let me know in comments below if you know, and I’ll update here once I’ve experimented and been able to tell.
Carta is the Gold Standard for startups to keep their CAP Table, but at a price.
One of my companies hasn’t really raised any money, but we have a 50+ stakeholders do to a merger and employee options. We execute maybe 2-3 documents per year related to capital. So the $8,400 annual price of Carta cost us about $4,000 per transaction that we did. Obviously, that is absurd.
We ended up downloading all of the reports and PDFs of all existing options. And added some instructions for what we need to do when new options are granted, exercised, etc. We save the CAP table and related documents in a Google Drive (that we already pay for), and ended up saving $8,400+ per year!
I understand that there are a few other things, such as 409A valuations and peace of mind that come with having a professional software like Carta manage your CAP table, but the savings, for us, are an easy trade-off.
The Net Promoter Score can be a pretty valuable metric for determining customer happiness, and, more importantly, how likely your customers are to tell other people about your product.
The basic idea is that you ask customers how likely they are to recommend your product to someone. Those who respond as a 9 or 10 are considered “Promoters”. When asked about your product, they’ll respond positively and encourage others to use your product as well. Customers who answer with a 7 or 8 are satisfied, but not likely to talk positively about your product. Customers who answer with a six or below are considered “detractors”. When asked about your product, they’ll respond negatively, detracting from your reputation. If you have a higher number of “promoters” than “detractors”, then your NPS Score will be positive. More detractors than promoters will result in a negative NPS score.
I was recently meeting with a leadership team and they mentioned that their Net Promoter Score was 6.6. That’s not a great score, but its not terrible. I don’t usually hear it expressed as a decimal, but I didn’t think much of it. After meeting with the team after several months, they kept mentioning NPS Score with a decimal and it had increased to 6.7. It was then that I began to ask questions into how they were calculating that. It turns out it was a simple average on a rating from 1 to 10. That is NOT an NPS Score! If anybody ever tells you their Net Promoter Score is between 1 and 10, make sure to dig in and make sure they are calculating it correctly! Scores should range from -100 (All detractors) to +100 (All promoters).
When calculated correctly, this product’s NPS score was actually negative. That helps to explain why revenue growth has been a challenge and marketing dollars are not moving the needle as they’d like.
Contrast that with another organization I meet with regularly. They calculate their NPS Score correctly and it’s a 60! No wonder this company has incredible growth and is doing well.
While your NPS score is negative, your first priority should be fixing the product and customer experience. Otherwise, every customer that signs up is likely going to detract from others using your product.
I’ve helped several teams lately go through an analysis of when to consider annual prepayments for services. These are some of the decision criteria and metrics that I use to consider if an annual contract or pre-payment should be considered.
As a baseline, calculate the full amount that you would pay monthly. For most software products, this is the regularly advertised price. Make sure you are looking at the actual monthly plan proce though. A lot of services have started advertising as “$x per month billed annually“. Make sure to select the monthly payment price whe you see that. Some services, like commercial insurance charge a small per-payment fee for “installment plans” that should be included.
Next, calculate the full price if paid up-front. Of course, you need to include discounts that are offered. Sometimes, an offer may make it a period other than one year, such as “buy now and get 13 months for the price of 12”, which makes it a little more complex. In that case, you could consider the annual price as 12/13 of the amount you pay. Or, if the extra month is not really material, you may chose to ignore the extea month.
After you’ve got those two numbers (the annual and monthly prices), you should consider the other terms and internal needs.
Consider if your usage of the service is expected to change much over the next 12 months.
Also, consider how much flexibilty you lose with an annual pre-payment. Some services, like Slack give you a credit if usage decreases. Others have no flexibility and you pay that amount, even if usage decreases or you cancel.
In general, I expect around a 15% discount for a full up-front payment and very flexible terms for changes in usage or cancellation. If terms are more strict, I’d aim for more like a 30% (or more) discount for the commitment and up-front payment.
Finally, consider your own cash flow and capital positions. If you have an plenty of cash in the bank, you can lean toward the saving of an annual prepayment. If you don’t have a lot of cash, You’ll favor the monthly terms.
What are your thoughts and experience? What else should be considered when evaluation annual payments?
I have multiple businesses, so I log into AWS multiple times per day.
That is a little tricky to do using LastPass since AWS has some hidden form fields that must be filled in
when using two-factor authentication through Google Authenticator.
In order to make it work correctly, I’ve had to modify the extra details in LastPass to add some extra hidden fields. If you set these up in your LastPass credentials for AWS, you should be able to log in with just a couple clicks, like usual, instead of having to type in some of those fields every time or having them overwritten.
Also, make sure to check the “Disable Autofill” checkbox an all of your AWS LastPass entries. Otherwise, one of them will overwrite the hidden form fields on the Two-Factor authentication page