Create Your Hotel Bill Book Format for Engineering Effortlessly

Hotel bill book format for Engineering

how do you design a hotel booking service [Music] okay so hotel booking system or service so let me think about uh functional requirements to start with so we are talking about hotel booking system so when when I talk about it is it like something like booking.com or like Airbnb or maybe deserving a small hotel or something yeah if you're familiar with hold like booking.com let's use that as an example for what we're trying to build sure so like booking.com there can be multiple hotels so there can be like maybe getting viewing a hotel page can be required then adding hotel maybe an administrator then deleting updating kind of those normal operations from the hotels then do we want to search hotels so maybe that's one of the requirement if you think the search then if if I have selected a room it can be I mean there can be multiple people booking a same room at the same time so so one thing is so are we going to be assigning room numbers up front or I think most of the hotels they assign room numbers at check-in yeah good question good question uh less for the sake of this exercise assume that rooms are assigned during check-in and we also do not want to double book um occupancy so just keep that in mind sure sure yeah those are most more about non-functional so uh requirements so mainly concurrency maybe not no double booking so if we have only one uh room available then only one person should get like for a particular room type I would say uh the room numbers are going to be assigned later so searching hotels is one option then maybe when we book a hotel there will be order created maybe reservation uh that can be one option for requirement maybe cancellations payments so what do you think I mean uh in terms of the functional requirements yeah this this this looks good um this looks good okay sounds good so then in terms of non-functional requirements uh let me see so and there are sometimes uh like hotels what they do is they do some kind of overbooking where uh where I mean if the like if there are even if the hotel is sold out they kind of book more rooms so that to prepare in terms of cancellations in future do we think do we want to support that feature or are we good without it um I'll I'll leave it up to you so maybe like initially we can leave it out but if you want to talk about it too I'm kind of curious how you would approach that so like okay so concurrency we discussed I'm pretty sure scalability to the extent of booking.com it's going to be highly scalable system and uh I I think from latency perspective I guess you should be doing hotel in general we make sure it is fast and I mean even if it is like 150 milliseconds or such kind of view and even search I think search should be fast in general so I guess that is could be one non-functional requirements and from security perspective yeah we want to make sure that uh people have like logged in they have their I mean the identity access management system is um properly built so even DDOS protection rate limited so those are standard things I think I would just assume that those kind of non-functional requirements would be good to get started uh yeah that sounds good that sounds good yeah I think this would be a good start for us to make sure this is a good user experience so that they're not you know users don't get frustrated and don't drop off in the experience and also from a performance um and security perspective right so yeah these non-function requirements sound good sure so in terms of backend the back of the estimation so there can be I mean uh let's assume there are probably one million no maybe hotels and uh each having some kind of 100 rooms so maybe 10 room types I'm just I'm just yeah throwing some numbers uh what looks like the decent um estimation so maybe 10 room types they can be king bed they can be like sweet or uh on an average maybe 10 room types of five room types I can assume so that comes out to be total of like 1 million hotels so around 100 million rooms so 100 million rooms are if we talk about room type so 100 million rooms if you talk about 5 million room types and if you're reserving like per day right so per day if you're reserving for maybe 10 years so that's going to be a lot of data to store so every day uh let me come up with the data model also and that will help Define how much storage it will need but uh yeah some it seems like it's going to be pretty good number in terms of like let's say if we are talking about reservations okay yeah and let's let's just do like a quick estimation here we don't need to get too deep here yeah one million hotels 100 rooms so 100 million per day into 365 45 equal to around one with 180 billion rules um yeah so so 180 billion rooms and uh that's going to be a lot of storage if you are storing every room reserved so even if it is taking 100 bytes per reservation so it's going to be probably 18 terabyte which is going to be huge so we need reservation we need some like we can't store it on one server for sure so the other thing I think here since we don't want double booking I think uh we would need some kind of inventory uh storage or so where we will have uh we will where we'll have like how many rooms are available how many rooms are uh kind of reserved so that kind of data storage in terms of table um we can have and we can surround it with different uh transactional features or other concurrency control mechanisms to make sure we are we are not double booking so here I think um here also let me first go over some data model and maybe I'll come up with those numbers sure yeah I think that'd be helpful yeah so but other thing is I think these hotels will have a lot of images at home so because you need when you are having viewing a king bed you might want to scroll through the images you might want to scroll through the images of the hotels so there will be plenty of images I think that can be stored in a maybe s3l come over in the data model and uh I think we'll need CDN coverage also that I'll come up with the design so even if it is 10 images I mean it's going to be huge storage even there so I think it cannot be stored in relational at all so let's see I mean if we talk about these system let me see here so we have we can have probably some databases or Services also I'll come up with so so maybe I mean uh let me see so if we talk about one hotel so one hotel can have Hotel ID then can have like its address or I mean those standard things that can be stored in there let's see if I can reduce down the font for this one so and other things latitude longitude we can store all of these things then if we talk about a hotel room so there can be Hotel there can be hotel rooms so I think so mostly it will be like room types so Hotel ID the then June type I would go with the inventory first let's see inventory so in eventually storage so for each room type for a given date how many uh how many are the available rooms for the given room type so available count and how many are the total room total room count for that particular day for the given room type so I think this will be helpful for viewing the inventory of a given hotel for a given a date when we select dates and multiple dates also when we select so that we can do in the inventory part and then let's see then people can make reservations so maybe some orders or reservation table we can have when when they are making reservation I mean for a given hotel which date I will need a start date and date and which rooms are being booked so room type kind of ID I would say ID here room type ID so we can have more tables for room type to room type room type ID to room type information and all those things but mainly for that particular room type ID how many rooms are being booked got it this makes sense like I I trust that you can do the same thing for a few other like you know hotels and other data models I'm more Curious now um how would you design this system for this I think everything up until here makes sense um just for the sake of time I wanted to speed through some of this sure sure so uh system wise let's see and I'll come up with some apis also then so here I think I'll create I'll have some Services okay one is inventory service there is talking to maybe this is talking to um we can build cash also but let me first have database here so like inventory service talking to this data storage so what I would do is if we have a user so let's say users so the user usually what I do is I mean for every page I try to make sure the like I build in a format where it's like back in for front-end format where if we have native like I If we have iOS Android mobile web desktop web or even if there is any other UI platform they can share the same back end uh so because the UI is a lot similar so that's why it reduced duplication so I usually have some kind of back and forth front-end experience or like experience layer uh there's mainly services and for this one I mean for any of the servers that I am going to have I will be having load balancer on top of it so like for if we are talking about a view hotel service so for View Hotel there is going to be a hotel page that will be powered through such kind of experience layer then we can have another something similar we can have like let's see here so if we are talking about viewing the hotel so for that if somebody is talking trying to get about the inventory for that hotel so they will call the inventory service also so that will get the database but there can be situations where I mean let's say if they make a payment so if we are talking about kind of preservation experience a reservation so then it will be talking to the reservation Service as well and it needs to talk to I mean when you do reservation it will need to talk to the inventory also to make sure we are having I mean we are having the right number of uh runes reserved so that both are in sync with each other let me go over that in a little bit sure so we'll have kind of reservation database so there can be a couple of approaches we can do some kind of microservice based architecture where reservation service has dedicated database um and they can be part of like the same database also but it will help is I mean if we are doing transactions and all then then having them on the same server would make life easier but practically if we think about bigger companies you usually have different teams that are managing the viewing the part like different teams that are managing viewing part or reservation part and there are plenty of complexities behind the scenes so a lot of the companies prefer having separate databases and separate storages that increases some complexity but let me go with that a little bit so in case let's say if somebody is making payments and payment fails so I'll have in reservation I'll have a status also where status can be probably I would say maybe pending if we want to support some kind of timeout I'll come over to that pending then complete maybe can be a straight then maybe cancel can be a state or so if let's say a reservation is canceled or so then we want to make sure that inventory is reverted also like available counts the ones that were reserved before that available account needs to increase because a reservation is canceled so for that I mean whenever we make payment a lot of the times the payment processor they reach out to us in an offline way through some kind of events processing so or callbacks so what I would do is maybe I'll create some kind of queuing mechanism as well so where I'll have some kind of queue event queue that and I'll create a consumer also so so whenever so this consumer will consume from those offline event queues and offline event queues will listen to uh some kind of events let's say payment failed so it will listen to I would say it's a payment failed some kind of topics if you talk about Kafka or rabbit mq those kind of uh like those cues do support listening to such kind of topics and events so payment failure or if it's probably I mean reservation canceled so in all those use cases I'm going to be as soon as those events happen I'm going to be listening to I'm going to consume it then I'm going to make the service calls appropriate service calls to for reservation service to make sure that it is canceled or the state is updated and I'm going to make call to inventory service also to update the inventory so I can go a little bit into inventory service but let me first uh uh let me first check this yeah and maybe or maybe we can kind of just walk through at a high level I think everything makes sense so far maybe walk through at a high level just from the user to booking a hotel like how like how all these services are working together just maybe in like a minute yeah yeah so when a user books a hotel they get the UI screen experience layer and they will be doing the orchestration uh through different services so when the user books the hotel it will call reservation service and it will also call inventory Service uh to inventory service what it will do is let me let me go with that one first okay so so the way I mean since I mentioned about inventory table here and reservation so if so the way it will do it is like maybe I mean such kind of queries like selecting first I'll get the available count I want let's say if there is only one room available so I'll first get the available count from this inventory table uh and let's say if I get this available count from the inventory table where if I have selected date between some start date and end date uh user provided let's say if I want to break from today through three days from now so in these I'll get the available count and I'll check if that available count available count is greater than what I am requesting requested count so or maybe less than requested count in that case I can't book the hotel because the rules that room is not available so for this one date Hotel IDL have uh equal to some Hotel ID and room type for given room type so so if the available count is less than the requested count then I can't book The Room in that case uh so and here in that case I would like to fail the this but uh in here I mean I get the available count and then I update this inventory table kind of so update this inventory table so Set uh I mean available count equal to available count minus the requested count so where where are all these things okay so what I see is like if two users are trying to book a last room so I can see some race condition sorry I'm going off track but I think this is an important one in this particular design so because we want to make sure it is not double booked right so if I'm going with a relational database so and I'm checking the inventory for the user here like two users are trying to book the same room so for both of the users the available count will be one and requested count will be uh one as well so if available count is uh less than the requested count right now I mean both of them will see available count is equal to requested count so they will proceed here and they will both update the inventory and reduce the available account so one of them can make the available count as negative so this race condition I think there can there are other there are multiple ways that we can protect it I mean we can protect it by adding a transaction on top of it so we can do begin transaction and end here but when we do this there are multiple transaction isolation methods like read committed read uncommitted and uh repeatable read and serializable so if we do anything other than serializable then we will still run into the same issues uh it's not so what serializable helps is it these transactions for one user it will make sure that this all these statements are executed once and the other user is waiting cannot uh like cannot proceed I mean because only one transaction can uh proceed in that serializable isolation level so in that transaction with only one transaction will proceed so for user one it will proceed and uh inventory will be updated correctly but for user 2 the available count would be zero so user two would be fine in this case the problem that I see over here is given that it is serializable we are blocking the transaction we are blocking the code for a long time and that can be inefficient potential of creating Deadlocks potential of creating having users wait for a long time to get the room available so it's not that uh I mean it it is good I mean it will solve the concurrency issue but it's not going to be that efficient so yeah so there are other ways I mean we can do select four update here uh it will lock the rows we don't need the transaction to be serializable there it will lock the rows it's still pessimistic locking there is other way where we can have a version in the in these like in these uh tables so it's like optimistic concurrency control uh that is helpful if there is not a lot of contention so and I think I'm just uh going a little bit faster but I think we please please interrupt if there are any questions but there is another thing that I can think of is having database constraints though it's not supported by all kinds of databases but the database constraints also aware I mean I can create a constraint um and and actually before like getting too deep into this I appreciate you walking through how you would optimize this race condition or how you would solve for it I think this year makes sense um for the sake of time and I think your overall system design makes sense here I would love to to discuss some other optimizations you would make so um race condition aside are there any other optimizations that you think could be made in your design so one thing is inventory I mean yes from the speed perspective like we mentioned it should be 150 milliseconds which is pretty high which is pretty like I would say strict uh in terms of uh the SLA so we can create some kind of cache here for inventory service where although I think the ultimately the booking part when you are ready to book ultimately it will go to the database and validates all these through constraints and make sure there is no double booking but when a user is saying they can see some kind of uh like that inventory service can talk to some kind of cache let me see if there is a there's no cash but yes I think we can have a block for cash that inventory service talk to mm-hmm okay uh with before talking to the database so that Cache can be built in a in multiple ways I mean that Cache can store the available count for a given room type A lot of the times I mean since people book well in advance maybe one month in advance and all those things so we wouldn't go get into the conflicts later on when it comes to the database but yeah so we can store the available count for a given room type in those caches cash and whenever it is booked either we can have we can update the cache offline uh to improve the booking performance uh we can remove the value like delete from the cash also so that next time when you read your reading from the database so if you delete the cache value I mean it will be slower next time so I would prefer that when we do a booking it goes through the same event queue mechanism uh like some kind of uh event that happens and that consumer will update the cash so that's one option one optimization then I was asking about images yeah yeah and and I I think that's great um now for the sake of time um I it's instead of diving diving into the same level of details for other optimizations can you just speak high level of like what are some other optimizations you might make if you had more time to work on the system design yeah I mean one thing is images I think I mentioned before so images will be stored in the S3 in S3 and uh that can be retrieved through CDN that is geographically located and um because the images take a lot of space and bandwidth as well so that's one option then other things [Applause] I mean not optimizations but there can be some like uh concurrency issues with when I talked about just inventory service but if we talk about updating inventory as well as reservation like those two tables at the same time because we what we want to make sure is as soon as you reserve a room that reservation entry should be there and inventory should be updated both of them should be at the same time so that one if we are going with the microservice kind of architecture I mean there are multiple ways like two phase commit is one option and uh there are there is something called Saga also so that helps in concurrency control and making sure that those two tables are in sync so that's that that's that will be helpful and even I mean if if you are double booking I think there are ways to optimize uh um ways to control the I mean make sure that people are not there are no two entries that are added to reservation table also uh for the same reservation but by smartly creating the IDS let me think about any other way to optimize this I think that's I I think that's that's fine for this exercise uh Niraj and I think we're at time now so um thank you for walking through how you would design this hotel booking service we talked about functional non-functional requirements you gave us an estimation we walked through like a high level design we went deeper there we talked about optimizing for uh solving for the race condition and optimizing for a few other things um now and I I think for for their purposes of an exercise or an interview this is the kind of details that we would get into now uh before we close out the video um given your years of experience doing system design do you have any tips for the audience if they were to face the same kind of question around booking a hotel system yeah I think most of the system design questions is one thing to remember is mainly we should follow the right format it's system design is just a kind of snapshot of what you will encounter in the real life so in real life if we have product requirements and we want to make sure that we are building a system that is tying to that those products that is adhering to the those product requirements so that's why we come up with functional requirements non-functional requirements and whatever system we are designing it's based on estimation if something does doesn't fit under the server then only we do I mean we can do sharding also here because uh that I couldn't get to but uh there are multiple ways but we have to make sure that we are solving for the problem in hand so that's why a lot of these steps are important high level is important optimizations are important and considering all these kind of use cases are important so definitely I think following the format uh and thinking through all the steps is definitely I would recommend for sure great great advice thanks Niraj and for the audience we're watching at home good luck I hope you found this helpful again these are typically asked in engineering or engineering manager typically in on-site or even in first round interviews so hopefully this is helpful and good luck in your upcoming system design interview thanks for watching don't forget to hit the like And subscribe buttons below to let us know that this video is valuable for you and of course check out hundreds more videos just like this at try exponent.com thanks for watching and good luck on your upcoming interview foreign [Music]